Day 1

03.09.2025

My September is all about talking to users. Today was day 1. I’ve never spent this much time “selling” before, and I’m a total noob at it. By the end of the month, I expect to have learned a thing or two and I thought I’d share my wins and learnings here as I go.

So what’s the goal?
I’m building in robotics. Long-term, I want to develop general-purpose manipulation models. But that’s a hard technical problem, and my background alone won’t get me a 100m+ pre-seed. I need to be smart: start with single-task, high-reliability applications that generate revenue while I refine data collection + fine-tuning recipes.

I don’t have anything to sell yet. But I want to prioritise talking to users, because building a demo in my office is very different from deploying an actual robot and I think I'm gonna get nerdsniped and will lose time. The only manufacturing shop floor I’ve seen so far was a gasket factory two years ago, and I need to see more of what manufacturing looks like in the real world, not the internet.

My goals for September:

1. Understand the real requirements manufacturers (or others) have for robotics.
2. Find one task common across multiple users → build a minimal prototype post-September → come back later this year and pre-sell the fully deployable version.
3. Build contacts that will be useful beyond the first task and develop an intuition for what this customer group is like and what they care about.

My main hypothesis I started with:
Start with contract manufacturers who have 100–1000 employees. They’re dynamic enough to make purchase decisions, usually deal with a high-mix / medium-volume setup, and might be open to conversations with someone like me.

First challenge: how to find companies to reach out to? @DominikEsen suggested using OAI’s Deep Research. Worked great, except it only returns 10 companies at a time. Pro: it summarises addresses, employee counts, websites, and phone numbers, perfect for planning visits.

I prompted for companies around Nuremberg (where I’ll be next week). What surprised me: how long it takes to understand what each company actually does and whether it’s worth contacting them. My girlfriend told me this is called “prospecting.” Makes sense. For now I’d rather be too specific and then learn which details are actually useful. So still a bit slow.

Tactics so far:
▶︎ Call first → usually reach reception, who then gives me the right email or tells me to write to info@.
▶︎ Write structured email templates. Took way longer than expected, but feels worth it.
▶︎ Draft a short, structured phone intro script after my first calls.

All in all, it took more time than I thought. I wish it were faster, but strangely, it’s motivating. Going to bed early so I can hit the phones again tomorrow during business hours.

Results day 1:
📞 Phone calls: 5 (4 picked up)
📧 Emails sent: 5
🤝 Meetings agreed so far: 0

Day 2

04.09.2025

Started the morning exploring some tools friends had recommended. Looked at four in total: three were more geared towards private equity (great for fund managers, overkill for me), the other was Woodpecker, which is more for email campaigns. Then I tried http://Apollo.io (thank you, Philipp from Telli). It’s really useful for detailed filtering of companies and has the emails of many employees. It also doubles as a CRM, but I’m just using it for prospecting for now. Two people told me that http://Clay.com might be better, but I haven’t looked into it yet. I also set up @Superhuman to see whether my emails are being opened, but partly also just to try it out.

A big question of the day: who do I reach out to - owners or people on the shop floor? I heard it’s easier to get authority for an idea if it’s decided on by the owner, so I’m going for that. At 50–200 employee companies, I expect them to be close to production themselves as well. One conversation got me worried though: the receptionist told me the CEO generally only books meetings 4 weeks out. Shit! Maybe I should have started earlier if I want to start meeting people in person next week - but fuck it, can't change it anyhow. I guess it's more of a motivation to keep going full speed.

A tactic I found useful today:
▶︎ I started a Google Maps list with all the companies I contact so I can see them on the map. Idea: when I’m on the road, I can stop by even if they didn’t reply to emails.
▶︎ The companies are often in industrial areas, and I started scrolling around to spot other nearby companies that could be worth reaching out to. Lower travel time if I can book two meetings close together, and if not, I’ve got more places on my Google list where I can just drop by.

@_sebastianscott replied to yesterday’s post and suggested getting a long email list and blasting people. This got me thinking whether my approach of clicking through company websites and personalising my emails is overly detailed and inefficient. It’s tempting and I want to keep it in mind, but for now I’d rather keep quality high + frequency low. Scaling to 100s of emails later is still an option - and maybe I’ll learn something useful while digging through websites.

One personal reflection:
Rejection feels very different this time. A year ago, sending cold emails felt hard. For what I’m working on right now, I believe so much in the tech and I’m convinced it’s essential for the future. I’m also 100% convinced of the value for manufacturers, so rejection doesn’t bother me. It feels like I’m trying to help them by telling them about this.

Day 2 results:
📞 Calls: 6 (4 picked up, 1 good conversation)
📧 Emails: 20 - better than yesterday, but I’d like to get to 50/day, not sure how yet.
🤝 LinkedIn outreaches: 2

Keeping track of the numbers on paper kept me motivated.

Day 3

05.09.2025

No emails today, but met a few people (not customers yet). One of them was @JannikGrothusen who kindly lent me his Trossen arm for the month. I’m still deciding on whether to buy an Arx arm or a Trossen, but wanted to bring a proper arm to customers to give them an idea of what the arms look like in real life.

Beyond the time I spent thinking about approaching customers in denser regions, which I posted about this morning, I also spent some time looking into Clay. It looks very powerful, but more targeted at bigger email campaigns. Out of the three emails I got replies to today, one was a more strongly customized email, which got me thinking that this might be the way to go for now.

Day 3 results:
📞 Calls: 0
📧 Emails: 0
🤝 Meetings agreed: 2 (first ones 🎉)

Day 4

06.09.2025

First visit at a company. This one manufactures and sells agricultural machinery. 3,300 employees worldwide and 1,200 in their HQ/main manufacturing site that I visited today.

1/ I was surprised by how little was automated. Probably because the machinery they produce is so big. I saw 3 robots. One was a specialized machine they co-developed with an external company over 3 years for ball bearings, which amortized itself across 8 years. The biggest system was a huge automated shelving system. Makes sense: 3rd-party system for a problem shared by many companies in a similar way. They also have a small robot that can autonomously pull trollies with items retrieved from the shelving system to the respective work station. 3rd robot is the UR. Didn’t have time to discuss, but will hopefully continue conversation on what it does next week.

2/ They have their own robotics software developer in-house, which I had heard to be uncommon. Probably more the case for smaller companies than for companies this size.

3/ I don’t think big machine manufacturers like this company are ideal customers for ML-based robotics: The components are big and mostly heavy metal parts that would require bigger robots. Also, volumes are very low. For the trailer that we discussed in more detail, they produce 7 a day (with 200 different configuration options!). There are also smaller components with higher volumes, like the ball bearings, but still not ideal.

🟡 Interesting note by the owner:
The German business runs at the highest efficiency worldwide (compared to Brazil, Eastern Europe, and the USA), which they attribute to 1) the regional supplier networks and 2) the skills of the average blue-collar worker due to Germany’s globally unique apprenticeship system.

Day 5

07.09.2025

Barely any work today. I sent 2 cold outreach emails in the evening.

I did, however, spend some time thinking about yesterday’s company visit and about entrepreneurship & persistence in the context of social media and geography.

In the early 1980s, the business I visited yesterday was started by its founder without an engineering background. For five years, the machines kept breaking down. For nearly two decades, the company barely grew, expanding beyond its initial workshop only after almost 20 years. Real growth came much later, in the 2010s.

Stories like this are emblematic of many German and European companies: beginnings marked by little knowledge, little expectation, and a lot of persistence.

That stands in stark contrast to today. Many founders now start with the expectation of building something big from the outset. Motivation often rests on scale, not just on making something work. It makes me wonder: how many of us would persist for years if we knew our company would never grow beyond a few dozen employees?

Part of the difference lies in culture. Social media constantly highlights overnight successes, hacks, and rapid scaling. Working on fragile prototypes for years with no team feels abnormal when compared to “0 to 100m ARR in X months” stories pushed on LinkedIn and X every day. At the same time, social media also provides proof and practical knowledge that fast success is possible. Is this a constructive push toward efficiency, or a disease of short-sightedness? Perhaps the hardest skill today is not execution itself, but the ability to think long-term in the face of constant noisy signals that suggest you are doing it wrong.

Another shift is in the role of networks and place. In the 1980s, whether you lived in a city or a village, your access to knowledge was similarly limited. Information spread slowly, and even if you had the right instincts, answers were hard to obtain. The internet has changed this: in principle, all knowledge is accessible everywhere. In practice, you still need to know what to look for, and informal knowledge still spreads faster through networks and in cities where connections are amplified by the internet. One founder might learn from friends that spending time in Shenzhen is critical for hardware speed; another, without the right circles, may never hear of it.

This makes me think that founding large, physical-product companies in small towns may now be harder than it once was. Information is no longer the bottleneck; networks are. For a country like Germany, the implication could be: if you want more great companies, you must not only fund research or infrastructure but also connect young people with promising ideas to the networks that amplify them. Without that, entrepreneurship risks concentrating only in cities, a trend that already seems visible in recent years.

This is more of a “thinking out loud” post. I’m sure I’m missing pieces, but writing it down helped structure my thoughts, and I’d be curious what others think.

Day 6

08.09.2025

Today was my first real customer visit from cold outreach. A 140-person plastics contract manufacturer for moulded parts. I had reached out to the CEO, who joined the meeting with a second employee.

I came prepared with two things:

1. iPad folder of demo videos showing how teleoperation works and examples of autonomous imitation learning.
2. The Trossen WidowX robot arm as a physical prop to spark imagination

Structure of the visit:
👉 I presented the tech and its limits (15 min)
👉 They shared their business + pain points (15 min)
👉 Open discussion of possible use cases (30 min)
👉 Factory tour (milling, deburring, attaching metal parts, roughening + painting) (~30 min)
👉 I thought the conversation was over, but we talked for another 30 minutes after returning from the tour

What stood out:
▶︎ They already use a Universal Robot, but reprogramming for small batch sizes (20–100 pcs/order) is too much effort → huge barrier
▶︎ Skilled labour shortage is the #1 pain point. When I asked if it takes months to hire two people, the CEO replied: “Months? It feels impossible right now.”
▶︎ He nearly archived my email, but replied because (a) it hit that nerve, and (b) I had clearly read their website and noted their robot
▶︎ He gave me tips: emphasise the labour shortage angle more in future outreach

6 Takeaways:
1/ Personalised emails work: showing I’d read the website and understood their context was decisive
2/ Shortage of skilled workers is the central wedge and should be front and centre in outreach
3/ Seeing actual tasks firsthand (milling, deburring, adding metal inserts, roughening + painting) builds instant intuition for what’s automatable
4/ People visit my website. It’s geared towards a technical audience at the moment and not optimal for sales. Building a good website is not a priority and makes more sense as soon as I actually have something to sell, but it’s still good to keep in mind.
5/ Main value prop: automate tedious steps so companies can get more done with the same headcount
6/ Personal takeaway: watching these processes live is key to building intuition for where robotics fits. The use cases I saw today are not ideal for the first application, but still super interesting.

We ended up talking for 2+ hours despite the meeting being scheduled for 1. In the end, it felt like the CEO was pitching their company as an interesting partner for product codevelops and brought up past cooperation cases that worked out. This was really cool, because it was a clear signal they’re eager to explore. (Too bad the problems I saw aren't ideal starting points)

Sample size is still n=1, yet this was a very very cool first conversation: it was so motivating to see technology could solve such an acute problem, but also to see how we got along well on a personal level.

📧New cold emails written today: 2

Day 7

09.09.2025

▶︎ 1 more reply today → phone call booked for tomorrow. Nice.
▶︎ Tweaked my email template to highlight the skilled-labour angle. Had second doubts further into the day but will see how it goes and maybe revert back to the old template.
▶︎ Tried mapping out companies via Apollo + Google Maps for companies in the same industrial district to reduce driving times. Rather slow though and not sure if this makes sense with a low response rate. Switched back to just working through the Apollo list only.
▶︎ Only doing calls now if the responsible person's phone number is listed on the website directly. It feels like calling the reception just leads to a "Yes, very interesting. Please send an email to info@..."
▶︎ Kept thinking about the meeting yesterday and how much better the tech needs to be to work for plastic moulding use cases. Doesn’t matter now. Emails first.
▶︎ Writing emails still feels slow and unscalable. Probably too early to judge → better to shut off the brain and just keep sending.

Today's output:
📩 24 emails
📞 1 cold call

Day 8

10.09.2025

I had two very interesting calls today, one with a bit of negative feedback, the other not with a potential user, but maybe even more interesting.

Call 1: Precision tools company
👉🏼 I had emailed the CEO, and he forwarded me to the Head of Innovation. He suggested a quick phone call.
👉🏼 He liked the technology and explained how at his old job (an OEM), they automated a full line, but programming took forever — so he liked the appeal of teaching a robot by demonstration.
👉🏼 Their production is the assembly of many small parts for power tools, with quantities from 5 to 6 digits a year.
👉🏼 He was mostly interested in a logistics case: a robot batching incoming supplier parts into preconfigured amounts for their fully automated small-parts warehouse.
👉🏼 I tried to push for an in-person visit, but he said he needs something to show in the next months, and that I seem nice but without a product I’m not interesting (he admitted he only took the call because he was driving and would otherwise listen to music).
👉🏼 I recommended @SereactAI for the warehouse problem since I spoke to Marc last week, who was super helpful. It was just a small, spontaneous way to return the favour (thanks @nathanbenaich for connecting).

Thoughts post-call:
A professional website + good demo video would have helped here, but I chose to talk to customers first instead of building in isolation for months — no regrets. Might change the website to a clean Framer template that looks more professional instead of my current one (screenshot below), which is a bit nerdy and was a simple Next.js page I put up just to not leave it blank (could no website actually work better for these kinds of customers?). I feel like a better website could definitely improve email response rates because it’s the first thing people do after reading the email (as shown in my web analytics).

Call 2: Ex-shopfloor guy, now building CNC tending machines
👉🏼 We talked for an hour. He said cold email didn’t work well for them — what did work was showing up in person (but not mornings 9–10, that’s when people talk about who’s sick, who fought, and which machine is broken). Afternoons, especially Fridays, were best. Conferences also worked, but only if you have something to show.
👉🏼 Because of his shopfloor + installation experience, he had seen many use cases. After watching my videos, he suggested ideas I added to my “Application Hypotheses Sheet.”
👉🏼 Most interesting: tending machines for O-rings or sealing rings. Imitation learning has an edge with flexible/compliant parts (like the classic t-shirt folding demos), and sealing rings fit that perfectly.
👉🏼 He also rekindled my interest in bin-picking — most use suction, but some companies prefer grippers (we’re still texting about why).
👉🏼 For smaller companies, packaging with varying object sizes can also be tough. He mentioned one company where two people pack small steel plates of varying sizes full-time.
👉🏼 Not sure if this is the ideal first use case, but overall the value of imitation learning is not that you get a robot specialised for one task, but one whose task you can change - so it doesn’t become obsolete.

Excited to try out showing up cold at the first places tomorrow!

Also continued doing some outreach today:
📩 Emails written today: 10
🤝🏼 LinkedIn requests: 1

Day 9

11.09.2025

Three things happened today:

1/ Wrote more emails
Last Thursday alone: 20 emails → 2 replies
Since then: 40 more emails → 0 replies

Not sure why. Without responses as feedback, motivation drops quickly. Today I tried changing things up:
1️⃣ Use the casual pronoun (‘Du’ vs ‘Sie’). In German, it stands out because nobody expects a mass mailing to use informal language.
2️⃣ Write hyper-personalised emails. That’s how I got Monday’s meeting: I mentioned their new UR arm and the pain of limited programmability - it clearly resonated. Another meeting from last week is scheduled for next Monday.

I want to reach 100 emails to make sure I’ve given this channel a proper attempt, then try retargeting and then focus on exploring other channels.

2/ Ordered business cards
I’ll be at EMO Hannover in 1.5 weeks, and business cards will be essential. My sister designed them (thanks! 😘).

3/ Tried showing up cold in person
Emails + cards in the morning, then I went to visit companies directly. Bad start: I left at 3:30. At the first office, the right person had just left; at the next five, nobody was at reception. I tried speaking to someone leaving the production hall, but after my second question he cut me off: “Please leave me alone and read our website.”

That was frustrating. Partly because I genuinely want to help these kinds of companies, and partly because I was annoyed with myself for not leaving earlier. I’d heard they close early, but thought that meant 6, maybe 5, PM, not 4 PM. My girlfriend (who’s from here) told me some people start at 5 AM to finish by 2 PM.

Lesson learned: leave the house way earlier tomorrow.

Emails sent today: 7

Day 10

12.09.2025

Today was my first day of doing cold in-person office walk-ins. I started at 10 AM and managed five visits before noon:

👉🏼 4 rejections (2x because the right person was in a meeting, 1x because the company was too large and reception blocked me, 1x no luck at all).
👉🏼 1 short but interesting conversation with a pneumatics company that also acts as a systems integrator.

That chat lasted 15 minutes (the guy had a meeting), but it was great: he showed me an actual use case: unboxing a basket of steel rings and placing them on a tray. The tricky part: moving aside the plastic foil on top of the rings, something traditional robotics struggles with. He mentioned the customer had a competing offer for €400k and that they’d use Keyence sensors, where the camera system alone costs €40k. He even asked for a business card. Well, good that I ordered them yesterday.

Afternoon round (7 more companies, 15–100 employees each)

👉🏼 1:30 PM: Owner and secretary already gone. Entered through the factory side door and chatted briefly with two CNC operators — interesting, but not very productive.
👉🏼 2:00 PM: Just caught the last employee leaving.
👉🏼 2:05 PM: Quick 5–10 min talk with an owner who simply said: “We don’t want to automate humans.” End of story.
👉🏼 2:30 PM: Secretary tried to find the owner with me, but he had already left for the day.
👉🏼 2:45 PM: Plastic moulding company. Spoke with a worker running CNC machines to make moulds. He liked the demos, but had no authority and said the colleagues who pour the plastic were already gone.

At this point, I was half-annoyed, half-amused. Everyone seems to go home before 3 PM, what the hell? I debated whether to drive to a lone factory building nearby (15 minutes, probably empty) but decided that if this is going to suck, I might as well feel the pain fully to remember it later and use it to motivate me when I'm building.

👉🏼 3:15 PM: Reception closed. But I spotted a delivery entrance, looked up the younger owner’s name (son of the founder), and went straight in with the robotic arm in hand. He was sceptical, but I could see him looking for Airbnbs on his computer so I could take the "I'm a bit busy at the moment serious at all". After two minutes of chatting he agreed to talk while dropping something off. On the walk, he opened up a bit more and, maybe because it was Friday afternoon, ended up giving me a short tour.

The factory produced plastic foils. Mostly moving roll cores and operating large machines. Not ideal for small robots, and the second plastic plant I’ve seen that makes plastic seem like the wrong vertical (same for CNC milling, though I’m still undecided there). Electronics feels more promising.

Still, I was proud I managed to convince him to show me around. On the way out, he mentioned a real pain point: new hires need 4–6 months to learn to operate all machines. Documentation takes too long; ideally they could just ask someone ... I told him about custom GPT models. He was really interested, so I promised to send a quick screen demo as a thank-you for him taking the time.

Might actually have some hopes for walk-ins! Friday afternoon might not be the best time after all though.

Evening

I went for a run, then had my second phone call with the owner of a 140-person electronics company in Poland (intro via a friend). He was super excited after the videos I’d sent. The call roughly went like this:

Him: “When can you visit?”
Me: “Friday in two weeks?”
Him: “Only one day?”
Me: “I don’t have a product yet, so it’d just be a visit.”
Him: “Ok, but do you want to do an internship at my factory?”
Me: “Internship?”
Him: “Yes, I’m exaggerating, I mean, do you want to do the job yourself for a bit?”
Me: “Ahh, yes! How long can I come for?”

So now we’re discussing details, but it looks like I might be spending 1.5 weeks in Poland soon, assembling electronics. Hell yeah!

Day 11 & 12

13-14.09.2025

I spent the last two days working on my next Substack post. Good chance to reflect on the past few months, what I’ve learned, what’s been frustrating, and what’s ahead. Most of that will be in the post I’ll send out mid-next week, so I won’t repeat it all here.

A more interesting moment was when a friend texted me asking how the week went. Answering that made me think about just how much of a rollercoaster it was.

👉🏼 Monday started strong: first meeting, the company loved the idea, and I felt real personal chemistry.
👉🏼 Tuesday–Friday sucked. Almost no replies, one phone call that went nowhere, and Thursday’s office visits felt like a real disillusionment.
👉🏼 Friday ended better, though: great talk with an electronics manufacturer in Poland. It’s now confirmed I’ll spend a week in their factory in two weeks’ time. That gave me a huge boost heading into the weekend.
👉🏼 Saturday night, I went to a dinner and, by chance, met the CFO of a battery recycling company. Still unaware, I explained what I’m working on, and he told me about their battery disassembly problem and invited me to visit their plant near Aachen. We talked about Europe - both our concerns and our optimism. That conversation reminded me why I’m doing this, and I left with a lot of energy.

The main reflection that came together this Sunday:
It doesn’t matter if 95% of attempts fail; what matters are the hits. A few weeks, months, and definitely years from now, I’ll barely remember the misses anyway. Every email or cold visit is just a lottery ticket. I can’t control the winning number, but I can buy more tickets. And by looking back at what worked and what didn’t, I can stack the odds in my favour over time.

I thought that the Substack after the one I’m writing now should be a numbers-driven evaluation of my outreach. It’s motivating to think that the more attempts I make this week and after, the more valuable that analysis will be.

I’m happy I started writing this very irregular Substack two years ago. The thoughts and ideas that come with every post are a pleasantly unanticipated side effect.

Very excited for this week. Time to buy some lottery tickets.

Day 13

15.09.2025

Today I had my second in-person visit that came out of cold email outreach 1.5 weeks ago. The company: ~400 employees, focused on steel punching. Most of their production goes into electric vehicle components, but they also run a contract manufacturing business with small to medium volumes. Their main value creation is building the punching machines, but they also operate them.

The big automotive lines are fully automated. The smaller machines, however, still require manual tending by humans and that’s what they‘d be interested in automating further.

Most of the tasks are pick-and-place, but not trivial at all at the speed they were running. One operator was simultaneously grabbing four rings coming out of a machine AND tending a different machine at the same time which required using a tool that could grab 10 smaller rings at once.

Technically, a traditional robot could handle these tasks because the parts always come out in the same spot. But the pain point is programming: if a line only runs for 1 day to 3 weeks, it’s simply not worth the effort. Not the best case for ML based robotics because it’s 1) repetition based with minimal flexibility being required and 2) reliability is essential.

I met directly with the CEO: mid-50s, physics background, and currently pushing for a regional robotics transfer center as part of an SME interest group he leads. I’m starting to notice a pattern: the two people replying to my outreach were already engaged in robotics and already aware of the problem with traditional robotics. It would be great if I could systematically find more of these people, since it could make campaigns much more efficient. But based on the two samples so far, I don’t see anything I could Google or filter for in Apollo.

Favourite moment of the meeting:
Me: “This robot I brought costs around 5000€, so…” CEO: “…basically nothing. That’s excellent!”

Most interesting number: ~50,000€
That’s the maximum budget the CEO said he’d currently consider for a first project together.

I spent the rest of the day driving down to the southwestern tip of Germany, where I’ve got another very interesting visit tomorrow.

Day 14

16.09.2025

Today I visited two very different but equally interesting companies.

First visit: Precision Steel Casting(@simtue‘s family business).
My first question: why cast if you can CNC mill?
Answer:
👉Cheaper: higher volumes, less expensive machinery.
👉Some shapes, especially with small cavities, can’t be milled.

How it works:

1. Mold wax to match the CAD shape
2. Glue wax pieces onto a tree
3. Dip in liquid clay, coat with sand, dry, and fire until hardened
4. Melt out the wax to leave a ceramic cavity
5. Pour in steel alloy, let it harden
6. Break off the ceramic to retrieve the part

Simon‘s a pretty sharp guy who had explored robotics himself before starting his current startup (he’s also doing daily updates atm!). He suggested three automation ideas:

1. Retrieving pressed wax shapes from molding machines
2. Glueing wax shapes onto trees
3. Automating final inspection

Live observation for (1) and (2):

• Many precise hand movements, sensitive wax, and tool use → not great for VLA robotics.
• Programmable robots could work, but with >1000 objects, programming costs outweigh labor.

The inspection idea, however, opened a new path:
👉 Every item is checked for blemishes anyway.
👉 A robot could sort baskets, photograph each part, and run anomaly detection.
👉 Moving parts tray-to-tray isn’t time-critical; tasks could run overnight.
👉 Parts are robust, but shapes and orientations vary. Perfect for VLAs learning.
👉 Over time, data collection would make the model generalise across new objects.

Even better:
If robots generally took pictures whenever they handled a part (doesn’t have to be an inspection task), production teams could trace errors through the process. Cutting material costs by even 1% could mean six-figure savings for SMEs.

This makes inspection/testing the second high-potential area for VLA robotics I’ve seen (the first being PCB testing in Poland).

Thank you for an incredible visit @simtue!

Visit 2: @SereactAI in Stuttgart
I’ve never seen so many robot brands running at once. I won’t share details to avoid confidentiality risks, but one thing’s certain: a very smart and very likeable team.

Thanks @TuscherMarc and @nathanbenaich for introducing us initially!

Day 15

17.09.2025

Started the day with a longer conversation and a run with Wh about staying focused on building research and technical know-how, without shying away from building for real customers.

Then: 9 cold visits → 0 wins. Rules learnt:
🚮<25 employees = no automation use case ⛔ 250 employees = reception fortress 🏰

Then: 4:30 PM, driving home, day basically done…
Spot a black Porsche w/ Swiss plates 🚘🇨🇭 Brain: 1️⃣ That’s gotta be the CEO still inside 2️⃣ Swiss? ETH Zurich might click 👀.

So I turned the car around for one last attempt. I actually thought of your story, @cornelius_schra, closing your first big deal by spontaneously stopping on the way to the airport.

I rang the bell, he opened, and we chatted. While the other 9 interactions that day had met with minimal interest, he was blown away by the iPad videos and how far robotics has come. I found myself just as excited: “I know, right?! Isn’t this so cool?”

He gladly showed me through production, already highly modernised with metal processing machines. The use cases he highlighted were mostly welding and setting rivets - again, not ideal for imitation-learning robotics, since those problems require easier programming for constant repetition, not adaptability on the fly.

Then we entered the assembly room, where we discussed 3 possible applications:
1️⃣ Pre-assembly: A robot positions objects so that a worker only needs to add screws (pictured).
2️⃣ Quality control (a repetition!): A robot sorts through assembled products and takes pictures to ensure quality before shipping.
3️⃣ Folding cardboard boxes for packaging.

Although there were no new perfect use cases, as I walked out into the sunset, I thought how cool it was that this spontaneous visit had transformed the day from one of pure rejection to a successful outcome. SO HAPPY I turned the car around. Never stop buying lottery tickets! What a mood booster for tomorrow ⚡

Made me think: I'm in St. Gallen Friday night to hold the opening speech of the Start Hack satellite event. Why not try doing some cold visits in Switzerland? I'm currently 20 km off the border and will exclusively focus on 🇨🇭 companies tomorrow.

Finally, I followed up by email with all the companies where the secretary hadn’t let me through:

📩 6 emails sent.

Day 16

18.09.2025

This morning started with a sting: a call was cancelled two minutes before the meeting, leaving me sitting in an empty meeting room for ten minutes. Salt in the wound for my email outreach.

Yesterday I finished the day writing follow-up emails and didn’t have time left to plan today’s route. I spent longer than expected planning. My focus was on Basel. New hypothesis: Swiss companies might be a good fit (world’s #1 in innovation, high wages, ETH name carries weight). Just as I was done and about to get into the car, I spotted a contract electronics manufacturer only 14 minutes north of where I had slept in Germany. A detour from Basel, but this matched my main ICP hypothesis.

I drove there. Two possible contacts were in a meeting, but I was proud that I even got past reception at a 1,000+ person company. It was 11:45. Lunch break coming up. Originally, I wanted to use that time to head toward Basel, but this company felt like too good a fit. The secretary was open to helping, so I gambled: had a coffee nearby, did more prospecting, and tried again after lunch - risking an hour of Basel time.

It paid off. I didn’t meet the original contacts, but I did speak to the head of automation. Over coffee, he told me about two problems:

• PCB inline testing
• Quality control of parts being delivered

Both things I’ve already heard before. Pattern? He also mentioned they’re considering a Keyence quality-control system for €40,000… which still requires a human to place the parts in front of it. Doing this with a simple camera + robot would be far cheaper and more scalable. Even just having a robot place objects in front of the Keyence would already create value. PCB testing (same project as the company in Poland im speaking too) is outsourced, so there’s money to be saved there too. Useful signal. Another call with the supervisor of the person next week. The gamble was worth it. 🤝

Afterwards, I crossed into the German side of Basel: three unsuccessful visits. At the last one, there happened to be an e-commerce company for pet supplements in the same building. Not my target market, but I was already there 🤷🏻‍♂️. I pitched packaging with two arms. They liked it. The right person wasn’t there, but we’re calling next week.

On the Swiss side of Basel I had two more tries left:

1. A 14-person electronics job shop. No time, but interested. ETH mention seemed to help. Call scheduled next week.
2. A fire electronics manufacturer (pictured). Shop floor manager unavailable, but instead I was shown around by a worker who loved the demos. The work was mostly assembly tasks, trickier but maybe doable in a year or two. I could even take a finished part and a set of individual components with me. Asouvenir and a good reminder of the dexterity needed. One part involves screws though, which might be tough. Currently, they are using a machine for parts of it. But it’s 38 years old (!) and falling apart. Still, the enthusiasm was great. I initially explained the concept to him outside over a Diet Coke while he smoked. Another coworker having a smoke joined for my “robot entertainment programme” and ended up touring me through the factory too. Having the physical arm with me made a big difference here (h/t @JannikGrothusen)

Wrapped up the day with two calls from a highway rest stop:

1. With a KIT engineering student who wants to get into robotics.
2. With @caluckenbach, swapping notes on cold sales (he’s done way more than me -Cal, you should do a post about this!).

Now I’m crashing at a friend’s place in Zurich. Already late, so route planning will happen tomorrow again. Better get some sleep now and wake up early.

Cold visits: 7
Successes: 3

Not that many visits, but what a ratio. Cold visits are already working way better than email.

Day 17

19.09.2025

No sales efforts today, in lieu of preparing a speech and attending a hackathon at my alma mater.

Day 18 & 19

20-21.09.2025

Spent most of the time working on my Substack. Again and again, I’m surprised by how much time writing takes.

Also talked to 4 other builders and 2 VCs. I’m currently keeping all non-sales calls for the weekends.

Day 20

22.09.2025

Nine out of ten cold company visits can be pure rejection, but sometimes you just meet an absolute legend. Today was such a day.

I started the day by setting goals for the week. Even though I’ll be at a conference in Hannover, I still want to hit at least 30 door visits. An improvement over last week: I began in the morning. The downside was that most responsible people were in meetings. Are cold visits maybe something that works particularly well in the afternoon? Not enough data yet.

I used lunchtime (when visits are pointless anyway) to drive from the Zurich area to Basel for a call with a friend of a friend working on microchips in Berlin. It didn’t lead to a direct use case, but I sent demo videos, and we‘ll stay in touch.

In Basel (on my way to Hannover), I visited three companies. Two were irrelevant because their products weren’t suitable for AI-based robotics. At the third, the owner was in a meeting, but it might still be relevant, so I followed up by email.

Then came the big visit of the day: a foundry producing cast parts from copper alloys, things like outdoor faucets, pipes, and fittings.

Normally, I have to push hard just to get a short chat past reception. This time was the opposite. The reception desk was empty except for a phone and an internal contact list. After some online background checks, I reached an operations manager who had been at the company for 15 years. Ten minutes later, he came down.

I started my pitch, ready to convince, but his third sentence was simply: “Shall we just grab a coffee and go to a meeting room?” On the way, he handed me goggles, a vest, and steel-capped shoes - asking only afterwards whether I even wanted to see production. (In my head: Hell yes I do, that’s the whole purpose of the trip!).

He was incredibly open from the getgo, sharing his experience, showing me around, and encouraging me take photos and videos (with the caveat that I can’t post them publicly, sorry 🥲). I showed him my demos, which he thought were “super cool,” and he explained use cases I only fully grasped once we were walking the production lines.

The tasks matched patterns I had seen before: machine tending, manual leak testing for the components (similar to PCB inline testing with green/red lights), and packaging.

Again, and I cannot stress this enough: watching the real processes being done and being in the production floor is so important to understand what the requirements for automation are. Visits like today make me so happy I’m doing this trip and remind me that it’s not something calls could come close too.

While they are not important insights two things stood out to me:

1. The sheer number of machines. Probably the most impressive collection I’ve seen so far.
2. The huge amount of custom machinery built for their top SKUs. The investment of time and money for custom-built, single-task machines is staggering.

I could also take some production items with me to work with as props back home. This ops manager is a legend!

By the end, I realized we’d been talking and walking for three straight hours. I didn’t hit my target of ten visits today, but who cares, successes like these are what I’m doing it for and those were 3 hours well spent.

I left very happy and drove eight hours north to Hannover for the conference. Unlike thus far, I don’t have friends to stay with here - but I came prepared: sleeping mat, sleeping bag, and a rental car with a big trunk.

A fantastic start to the week. Lots of energy for tomorrow.

Day 21

23.09.2025

Today was my first EMO, the world’s largest trade show for metalworking and machine tools. Conferences are one of the three channels I’m testing to talk to users (besides cold emails and walk-in visits).

The scale is insane: 90,000 attendees, more machinery than I’ve ever seen in one place, easily over a billion euros worth on display. At first, I thought it might be the wrong conference for me; most booths were CNC machine builders, whereas I want to talk to their customers. But an hour in, I realised it could actually be perfect: the real value is in approaching the people walking the aisles.

I used the first day to soak things up: learning about the machinery in use, the general vibe of the industry, and the type of people attending. The first booth I visited was an American oil producer selling liquids for CNC milling. I didn’t know you could build a 100-person company just around that … until I saw another 10 companies in the next 3 hours doing exactly the same.

Lots of robotic arms were on display, but without exception, they ran on hard-coded machine paths, zero intelligence. Which is good for me, though not surprising. The closest thing to “smart” robotics was a talk from the German CEO of Google’s Intrinsic AI.

Walking further, I found other halls with a broader mix: less pure CNC, more tooling companies, i.e. potential customers. The problem is, most were only represented by sales teams. Not ideal since I want to sell them something. Still, useful: I can now reference those conversations in emails (“I met your colleague X at EMO, who suggested I reach out…”).

I’ll do a full write-up of my most interesting learnings and impressions from the conference tomorrow.

Another unexpected cold email result
Yesterday, I got a reply to one of my cold emails from 1.5 weeks ago. We jumped on a 15-minute call, which I took in a quiet corner of the trade show.

I pitched for 5–8 minutes. Then silence (5 seconds that felt like forever) before he said: “Mr. Paul, everything you just described matches our problem 99%. Every few years we revisit automation, but give up after 6 months because of programming complexity and the product variance [shapes] we deal with.”

He then talked for another 5 minutes about their problems and use cases. I took the gamble and suggested an in-person visit in two weeks. He immediately agreed and even proposed a 2–3 hour workshop, where they’ll walk me through their production line.

Fuck yeah 💯 Sales is fun when it works like this 😂

One more call
I also had a follow-up call with a company from a cold in-person visit. Last time, I caught the production manager for just 5 minutes before his meeting. No immediate result yet, but a good warm-up for when I reach out again with a proper product in a few months.

Second night in the car
I went to the local John Reed gym after the trade show today for an exercise and most of all a shower. Now getting ready for the second night sleeping in the car. Writing these updates from an Autobahn rest area definitely is a vibe.

Day 22

24.09.2025

Day 2 at EMO
Yesterday was about exploring and catching the vibe. Today I came with the intent: to talk to users. Exhibitors were selling machines, tooling, and accessories. My target: the people walking around.

My approach
I focused on the robotics hall (Fanuc, Kuka). Their demos attracted crowds, so I used that as a weak filter for people who might care about automation.

I tried one of two lines:
Ask them about their work, wait until they ask back Open with “Want to see something these robots can’t do?”

Why didn’t it work
Relevance: Hard to judge who was relevant. Badges were small, company names vague, titles often unhelpful. No clear intent: With cold emails, my goal is a call; with walk-ins, it’s a factory visit. But here? A call made no sense, and asking for a site visit was too much. Without 3–5 questions, I couldn’t even judge relevance, and most weren’t from production anyway.

Results
▶︎ 3 business cards, where I can follow up later
▶︎ 3 semi-relevant conversations where I at least got a name to reference
▶︎ 1 jackpot that wasn't: one business owner loved it so much he introduced me to his entire team… but they’re in Turkey, not realistic for a development partnership.

Overall, it felt like I was just bothering people. I believe 100% in what I’m building, and with targeted outreach I know I can help customers. But here it didn’t feel right. 12 attempts may not be enough to dismiss trade shows as a channel, but the strategy was clearly off. I’ll need a better approach for MOTEK in two weeks. Ideas welcome.

Bright spots today
1️⃣ Met @HristakievD, who runs a 30-person team building bespoke industrial automation. Very cool.
2️⃣ Met the founders of Sightwise, spun out of PhD research, building defect-detection systems. Quality control keeps coming up as a major use case. Perfect for me: I don’t have to build it myself, and all their customers are potential customers for me.
3️⃣ Learned about €100k German government grants for SMEs applying AI and robotics. Opportunity (?): could pitch this to development partners as a way to cover costs.
4️⃣ Met with @jannikwh, whom I ideated with last year and who now co-founded @supplyco_ai in NYC, building a platform for manufacturers to identify ideal customers to grow sales. Jannik texted after he saw yesterday's X post. Lovely serendipity.

Takeaway
Two days at EMO gave me a good sense of the metalworking industry and how trade shows work. Not much direct customer progress, but valuable context, connections, and ideas. If I come back in 2027, I’d get a startup booth (~€3k) and exhibit directly. That feels like the right way to reach the relevant customers.

EMO goes on until Friday, but I’ll spend the next two days using the rental car for cold walk-ins.

Day 23

25.09.2025

Yesterday was my third night sleeping in the car, and I felt the effects: short, poor-quality sleep. The first two days, I didn't bother because it was great to be up for the conference early, but today I tried to catch up. Sleeping in until 9:30 still felt like I was waking up at 5 AM.

The lack of a morning shower doesn’t help. I’ve used the gym the last two evenings, but it’s too far from my night parking spot to drive there just for a shower in the morning. Working from the driver’s seat, though, has a surprisingly nice vibe with the music on the car's stereo.

I started with some brief Substack edits and route planning. Around Hannover, there’s clearly less industry than in Bavaria or Baden-Württemberg.

Six visits today:
3x: the right person was in a meeting or absent
2x: blocked by the secretary despite all my charm
1x: secretary tried to help but came back with a “not interested”

Not a great hit rate, but whatever, I understand by now that I just need to keep trying. The absence of luck today adds motivation to push tomorrow before the weekend. In the evening I drove to Berlin, where I can stay with someone and finally sleep properly. Tomorrow I’ll target companies in the Berlin/Brandenburg area before returning the car. Next week: Poland.

Day 24

26.09.2025

Big electronics company call split in two
Follow-up from a cold walk-in. By chance, I met the head of automation (1000+ employees) while talking to the secretary. What I thought would be a long 30 minutes turned into him sharing details of their automation projects. We continued later the same day, and he showed me plans for a new plant: 6 arms, reducing 20 manual jobs to 1 supervisor, producing 200k+ displays. Traditional robotics in action. When I brought up AI-based automation, he was interested but unsure about use cases - a pattern I keep seeing. We’ll reconnect in 3 months. If it works out (VERY big if), then this could be the kind of company that ends up buying 20+ arms.

One stop, technically successful
Berlin has more manufacturing clusters than I expected. At one site, the secretary turned me away, but I found an open shop floor door and was introduced to procurement. This meant sneaking past the secretary again, but luckily, she was distracted. We toured production: they make ~600 robotic laser heads a year. Interesting, though volumes were too low for a good use case. Still, a useful visit.

Small electronics shop call
Later, I spoke with the owner of a 12-person electronics job shop I had visited last week. He was very engaged, and we discussed cable work - tinning and crimping ends, 5k–10k/year. Small tasks overall, about 2 weeks of labour a year, so not really ideal, but he saw it as a chance to get accustomed to new technology. I suggested a system price around CHF 20k, concerned it might be high, but he replied: “That’s basically three monthly salaries - not too expensive.” Gotta love Switzerland. Cable handling is a cool case: practical for AI robotics, good to demo because everyone knows it and sees it as real work, and straightforward to work with in the lab.

The picture was my setup for the last call. So happy that virtual backgrounds exist.

Day 25 & 26

27-28.09.2025

No work for the company this weekend. I spent the weekend doing my taxes for 2024.

Day 27

29.09.2025

1️⃣ CRM:
Restructured my outreach sheet (merged cold visits + email overview). Still need to move cold email notes from Notion → don’t recommend Notion as CRM (no exports, no stats). Also wrote 2 follow up emails, many more still outstanding.

2️⃣ Calls (5 total):
👉2x potential customers Cold email → in-person visit next week. Follow-up to a cold visit (secretary blocked me) → positive chat, liked the tech (“I wish I could be working on what you’re building”), but the company is large (measuring instruments). He’ll forward my blurb + videos internally.
👉 2x exploration (thanks to @IlirAliu_‘s shout-out)
 – Aerospace founder (US).
 – Austrian founder → offered intros to 2 people.
👉 1x with a friend (ex-fund, now angel scout) → open discussion on what kind of structure might make sense for me in the coming months.

3️⃣ 6h drive to Płock, Poland
With the owner of an electronics company, I’ll be with this week. 6h in a car was a good chance to dive into his and the company’s background and get to know each other better.

Day 28

30.09.2025

Today was the first full day at the emergency lighting factory in Płock. We started at 8 AM and toured the current building and all workstations. The tour took 1.5 hours. What’s impressive is that for a company of their size they vertically integrate a lot.

They produce their PCBs in-house. First, a machine is used to apply the smaller electronic parts, which are then soldered by reflow soldering. Larger components are added by hand and fixed using manual soldering.

Afterwards comes programming of the PCBs, followed by in-circuit testing (ICT) as inline quality control.

In a separate process, they prepare cables: tinning the ends, crimping contacts, assembling them into cable harnesses, and pinning them into connectors.

Another line handles stamping sheet metal parts, which are then bent. Threaded inserts are pressed into the sheet metal.

The battery and the plastic housing are produced externally. Final assembly is then done in-house, with end-of-line testing.

The visit was very insightful. Watching the precise hand movements made it much easier to understand the process. Production volumes are in the low- to mid-thousands per SKU, which is an ideal range for AI-based robotic automation.

We also visited their new production site, which is more than three times larger. Unlike many of the other companies I’ve visited, they are actively planning to grow over the next two years, and therefore see automation as a way to scale without increasing headcount.

In the afternoon, I sketched the process in FigJam and used colours to mark my hypotheses about automation feasibility: good chances, possible but difficult, or very hard. Based on this, I‘ve prioritised which jobs I‘ll be doing in the next few days

In the evening, I exercised with the owner and his son, and afterwards we had dinner together, where we briefly discussed other AI tools that might be relevant for them to use (@langdock_hq, Doinstruct, @roboflow, Deltia, Nano Banana for product shots). Nice reminder how so many things we take for granted in our bubble is unknown yet very useful to others.

I’m not sure what I can share, so unfortunately, no pictures yet.

Day 29

01.10.2025

Day starts 6:20 to be in the factory by 7. We started with a short tour, then I joined the owner’s son to program and test PCBs. The workflow: place a board in a cassette, press start, wait ~90 seconds. Simple in theory, but doing it revealed quirks.

One example: I had assumed the blinking sidelight would clearly indicate status. In reality, it freezes in between for ~20 seconds on a random colour (green or red), identical to the finished state. For a robot, that’s a real ambiguity: short-term vision memory won’t (common in most VLAs) won’t work. You’d either need to tap into the machine’s computer directly or read its display. I prefer the latter. I don’t want any bespoke programming or customisations, but build a somewhat standardised product.

Another observation: I expected one worker to get through five boxes per shift after my initial phone call with the owner. After two hours we’d barely done half a box. Waiting time dominates. Ironically, that’s good news: a robot could run unattended for long stretches before a human has to swap boxes.

While labelling the tested PCBs, I realised stickering is also a strong automation candidate. Initially, I thought it would be difficult, but it’s actually well-suited for a bimanual robot: low risk, annoying for humans, easily verifiable by sight, and very common across factories. If a robot could master precise stickering, that would be extremely valuable.

After the (very short) 15-minute Polish lunch break, we switched to another PCB type with a different cassette. This one involved cable confectioning: attaching cable sets to boards. Each took me 50–90 seconds. Tedious, delicate, but again low-risk and easy to check. A perfect candidate for robotic learning, and widespread across companies.

We worked until 3, then I caught up on emails, went for a run, and spoke with @MaxAhrens in the evening. Thank you @jannikwh for connecting us.

Great second day at the factory - lots of promising tasks for automation, and more to explore tomorrow. Very happy I’m doing all of this by hand at the moment.

Day 30

02.10.2025

Day 3 working in the electronics factory in Płock. Today, working 4 tasks.

Metal drilling
We started again at 7 a.m., this time in the metal workshop with the owner. The first task was drilling holes into stamped metal plates. The setup wasn’t a full CNC, but a semi-automated drill with placeholders that made positioning straightforward. With square plates, this is simple: fix the plate, press the button, and the job is done. Round plates are trickier. They’re pinned off-centre and can rotate, so a robot would first need to align them correctly. For automation, square plates look promising since precision demands are low and motions are simple, though picking plates reliably from a stack would still be difficult.

Cable processing
Next was cable processing, where three or four women worked at different stations: cutting cables, crimping metal tips, and inserting cables into plastic housings. The insertion step is subtle; there’s a “click” when the cable locks, and a robot would need force or position sensing to verify the fit. Later, they brought out a different crimping machine, almost foolproof compared to the one usually in use, which would be impossible for a robot. The foolproof one only comes out seven to ten days a year for special contracts, but it highlighted exactly the use case I’m interested in: if a robot can be configured by demonstration in a day, it doesn’t matter if the task is seasonal or sporadic. You replay the program whenever needed, creating value even if the task is only repeated a few days a year. That’s the core of my current hypothesis: once you remove the cost of programming, robots become commercially viable for short-run, irregular tasks.

Cable tinning
After a quick Polish lunch break (15 minutes is common here) I moved to tinning cable tips. It’s tedious for humans but also not a good candidate for robots. The cable has to be dipped at exactly the right depth: too far and the plastic melts, too shallow and it doesn’t stick. Humans are both faster and more reliable.

Metal bending
Later I returned to the metal workshop, this time for bending. Stamped metal sheets are positioned in the press, the operator triggers it with a foot pedal, and after a few seconds, the part is ready. I could see how a robot could do this too: hold the sheet, trigger the press with its hand, wait, then turn or place the part on a table. The real challenge would again be at the start: reliably separating single sheets from a stack. Even experienced operators occasionally lift two at once because they stick together.

Evening conversations
In the evening, I joined the owner and his son for another workout, which was a good way to connect outside the shop floor. Afterward,s we had dinner at their home, where the owner and I spoke for three hours about my past AI projects, his views on robotics, how the company tracks and protects data, which trade fairs he recommends, how he recommends approaching other customers like him, a timeline for co-development and even a possible investment by his company. Spending this much time with him has been invaluable.

One comment stayed with me: as someone who started his business before Poland joined the EU, he is still deeply grateful for accession and the European project because it eliminated customs bureaucracy and helped his company grow. It’s the kind of perspective you rarely see in the papers, but one that shows how European integration tangibly shaped a small manufacturer. 🇪🇺

Tomorrow is my last day here, and I’m again very much looking forward to it.

Day 31

03.10.2025

Metalworking & Configuration
Today was the fourth and final day at the factory. We began in the metal workshop, bending sheet metal into shape. The operator showed us how to configure the machine and adjust its settings, which was fascinating. The feasibility of automation here depends heavily on part geometry. Since the pieces vary constantly, it’s actually a strong case for a flexible robot: high-mix, low to medium volume and errors aren’t costly.

One challenge: during bending, the robot has to release the piece. Ideally, you’d want a camera at the height of the press bridge, best positioned between the arms of a bimanual robot. But if the platform has an adjustable height then the arms would need to operate at different heights, which complicates design (I think?). The form factor of the robot will be an interesting question, but if I get to the point where that is my biggest problem, then I’ll already be in a good spot.

What made me smile was a comment from the operator. When we asked how he thinks about configuring the machine vs running it, he smiled: “I love the configuration, because it makes me think. Running the task is fine, but it gets monotonous, especially if it’s a job running for days, so I try to take breaks by configuring other machines in between.” That was really cool hearing: Robotics should empower skilled operators like him to focus on configuration, work that requires judgement and pays for mastery in skill, not repetition. This morning I learnt that even with perfect drawings, you still need experience to configure a machine optimally, and I don’t think AI will crack that in the next years (good looking trying to find a programmatic interface for that machine).

Powder Coating
Next, we looked at powder coating. Parts are hung on racks and rolled into a chamber, where they’re sprayed. For large parts, the process is already fast and doesn’t need robots. For small ones, especially when screws need covering, hanging the pieces on the rack is cumbersome and worth automating.

Assembly: so many diverse subtasks
In assembly, we built the lights. Assembly feels like one of the hardest domains: diverse subtasks like attaching and clipping zip ties, screwing, stickering, and inserting cables into PCBs. For example, screwing a PCB into a plastic housing was tricky because the screw hole was partly hidden. You rely on haptic feedback to sense alignment. That’s extremely tough for robots, especially with no good camera angle.

Packaging as a Market
Packaging looked simpler: mostly pick-and-place, folding cardboard, and some stickering. Still, dexterity is needed to push cardboard tabs into slots. But the market is huge. If you can automate packaging reliably, many small businesses will pay - shipping often consumes their first employee’s full time. A lot of companies like Hive exist because small businesses don’t want to do shipping themselves. @ultraroboticsco is doing something like this, but focusing on a part of it.

Car Ride Back
We wrapped up around 15:30 and drove five hours back to Berlin. In the car, we talked about hardware setups, costs, and even drifted into the philosophy of how neural nets work. Always fun to zoom out. I also showed @jannikgrothusen’s arm to the owner, and he was very impressed: He said we need way more people with Jannick’s mindset, who think they can build things better than incumbents, don’t let themselves be stopped and just build a better version.

Reflections
Four weeks ago, I thought I needed a polished, impressive-looking arm to win customers. Now I see that real connections and fast iteration on cheaper hardware are far more valuable. Money is nice - and last week I felt how even the idea of raising money can give you a dopamine rush - but intimacy with customers like this is worth so much more to build something valuable.

This week was probably the most insightful, productive, and fun of the last month.

Day 32 & 33

04-05.10.2025

Light weekend - mostly catching up on emails and messages. I went to Oktoberfest, then on Sunday evening drove from Munich to a town near Heilbronn for tomorrow’s 10 AM meeting.

On the drive, I listened to the first four hours of "Breakneck: China’s Quest to Engineer the Future". The author explains how China is an engineer-led country, while the US is lawyer-led. I especially enjoyed the descriptions and anecdotes about how China’s manufacturing hubs operate. Excited to go to China later this year/early next year.

Day 34

06.10.2025

Battery Pack Visit: New Use Cases & Quality Assurance Once More Confirmed
Started the day with an appointment at a company producing battery packs. What was really great was that three people joined the meeting. They had shared the videos I’d sent via email internally, and the two production managers were genuinely excited for the meeting. One person was responsible for innovation projects, two were from production, and they guided me through their line afterwards.

It was interesting to see the quality assurance use case again, but also a few new ones - such as preparing battery packs or placing cables into a rubber ring. The meeting put me in a really good mood; they clearly shared my vision of how robotics should work and why it matters for Europe. They also gave me a few battery samples and one of the cases they’re packed into, which means I can now replicate their setup at home and send them a demo. That should make it possible to talk about a concrete deal as a next step.

Two More Cold Visits
Afterwards, I did two more cold visits. In one case, I didn’t get past reception because the person in charge was busy. In the other, I met with the head of ecosystem and innovation. In theory, that sounded great - she was open and friendly - but the conversation didn’t have much signal. It felt like she enjoyed talking about innovation rather than doing it. She seemed quite detached from production and actual implementation, and couldn’t really say whether it would be useful, yet also seemed confident that it wouldn’t be. Overall, it felt like a bit of a waste of time. Still, she pointed me toward another company working more closely with robotics, so we’ll see if that lead turns into something useful.

Later, I went for a run and ended the evening with three calls with other founders working in robotics.

I’m starting to feel a certain saturation from all the user conversations.
Especially after last week in Poland, I feel like I now have a good sense of what’s valuable, what isn’t, how open people are to talking about robotics, and what helps get a conversation started. I’m happy to have a few good meetings lined up this week, but I’m definitely ready to switch gears and get back to building. In that sense, it’s quite fitting that this is the final week of Sales September. I think I’m going to reduce the number of cold visits and rather spend more time consolidating learnings and creating an overview of all use cases and impressions.

Day 35

07.10.2025

Started the day by writing a few follow-up emails, organising my outreach sheet, and structuring my thoughts on how to continue next. Then drove to Luxembourg to visit a company focused on recycling electric vehicle batteries, a fascinating and very complex problem.

They have already automated parts of the process such as unscrewing and cutting cables, but each battery model is different. There are hundreds of battery types depending on the car manufacturer, and even within the same model, individual batteries can vary. For example, if one has been repaired before and someone used extra glue when sealing it, even opening the casing becomes unpredictable.

What I had not realised until recently is that recycling EV batteries is so difficult because there is always some residual charge left inside. The cells themselves are mostly standardised, but the battery management system, which controls charging and discharging, is proprietary software. Every car maker uses its own, so recycling companies cannot access it. Without that, they cannot safely discharge the pack in one go.

As a result, every pack has to be disassembled manually, screw by screw and cable by cable, and the individual cells discharged afterwards. The process is dangerous. In some facilities, an electrician works on the pack while another person stands nearby, holding a rope tied around them. If the first person gets electrocuted, the other can pull them away from the battery to prevent serious injury.

There is enormous value in automating even parts of this process. But to do that, you need to master dozens of subtasks and, above all, handle the variability between batteries, from how strongly a casing is glued to how cables are arranged or intertwined. The company I visited had already automated several of these steps, but the remaining ones are exactly those where adaptability is required.

Once again, I realised how important it is to see the use case rather than just talk about it. During the first 30 minutes in the conference room, I did not fully grasp where the real challenges were. But once we looked at an open battery pack, everything became obvious, especially for tasks that sound simple in theory like opening a casing or cutting cables. Many of these involve subtle, force-based operations such as wedging open glued parts, something a human can do intuitively but that is extremely hard to capture in a fixed robot trajectory.

Ended the day driving back from Luxembourg to Karlsruhe, where I will continue tomorrow.

Day 36

08.10.2025

Today I did six cold visits, five of them at a technology campus near Karlsruhe, which turned out to be very efficient because of the high density of companies. I also stopped by the Daedalus office, but the most exciting meeting was with an electronics job shop of around 40–50 employees.

We discussed several tasks, glueing THT components onto PCBs with silicon, inline PCB testing, placing PCBs in front of a camera system, and separating partially punched-out PCB panels from the rest of the board using a machine.

I left with a very good feeling and an even stronger conviction that electronics manufacturing is the right direction to focus on.

Day 37 (Final Day)

09.10.2025

Today I visited a company producing lighting systems for buildings. We’d first connected two weeks ago through a cold email, which led to this site visit.

For larger batches, they already use a custom-built machine that handles long light strips in different configurations and even inserts cables into holes, though not without occasional errors. That system alone cost around €2 million. They also have a large automated machine for assembling the aluminium mount that’s fixed to the ceiling, with the lamp itself fitting into that mount. Together with partial automation in the electronics assembly, most high-volume steps are already covered.

Where manual work remains is in the smaller-batch lights, they apparently produce roughly 10,000 models, and for many of these the quantities are only in the hundreds or sometimes just one. Those manual assembly steps reminded me of the electronics factory in Poland last week: mounting, wiring, testing. These could, in principle, be automated with a robotic arm and a VLA-based system, but it would require strong generalisation across product variants to make it work. Not an ideal first use case.

The company is also currently using Kurzarbeit (a reduced working-hours scheme), so labour shortages aren’t a pressing issue. I felt like the guy I was talking with wasn’t the most cheerful person, maybe because the overall vibe of the company isn’t that great.

One interesting thing I saw was yet another version of testing, again PCBs, this time with a different technique called fly-over testing: you connect the board printed with LEDs to electricity and then check if all lights are working. You‘d need a suiting camera system though so unless I want develop that myself I‘d need to find somebody already doing that and sell it as a bundle. The company I visited has 0.25-0.5 FTE equivalents doing this task, so by itself not enough to warrant a purchase.

Afterwards, I stopped by the Sereact office and talked with Marc and then drove back to Munich, where I’ll return the car tomorrow.

That officially concludes Sales September.

I’ll write a full review on Substack, but overall, I’m extremely happy I did this. I saw dozens of use cases, spoke with countless people, and most of all, I feel deeply motivated. I probably got ten times more no’s than yes’s, but the yes’s were powerful and genuine. Meeting entrepreneurs who truly want to move things forward made me optimistic about automation in Europe.

Starting tomorrow, I’ll take some time to reflect on next steps and share my summary on Substack.

Thanks to everyone who followed along, replied, or reached out, many great conversations came from it.

If you ever get the chance, I can only recommend doing a sprint like this yourself.

— Dominique