Pinned Tweet
ACCELERATION IS NON NEGOTIABLE.
Brought back some posters from @prototype_cap event in Berlin, and this is the one my kid picked to put on his bedroom wall.
Brought back some posters from @prototype_cap event in Berlin, and this is the one my kid picked to put on his bedroom wall.
10
3
53
1,748
Pedro Oliveira retweeted
12 years ago I founded Boom, as a software engineer doing hardware for the first time. I’m probably the least handy founder in hard tech. My 12 y/o son does my soldering and 3D printing for home projects, and I’ve struggled with imposter syndrome the whole way through. What business does a software guy have building supersonic jets and turbines?
More than a decade in, the most insane realization is that software is the key to reaccelerating hardware engineering. But in a way that I don’t think is widely understood or implemented, even in the burgeoning hard tech community.
Most hardware engineering is actually software engineering in disguise. It’s spreadsheet “code,” engineering calculations done as one-offs. Eg wing structural loads in, spar sizing design out. Thermal loads in, air conditioning system sizing out.
Everything is a calculation, and calculation is software. But these calculations are done essentially by hand, as if we are doing arithmetic but have not yet discovered algebra.
Hardware engineers have baby software, built in spreadsheets, the kind of crappy software I wrote as a kid before I understood architecture, unit tests, APIs, version control, and everything else that makes modern software good.
Our airplanes and engines are built with the software sophistication of a script kiddie.
So what we tell our engineers: don’t design the hardware, design the *algorithm* for designing the hardware. And take the algorithm seriously—as an algorithm and as a first-class engineered software system.
Never design a piece of hardware, such that you couldn’t redesign it tomorrow with a different set of assumptions.
This all sounds obvious, yet our experience at Boom is that it changes everything.
Ten years ago we couldn’t afford to build much software. We had a small team—and there were few software engineers with the curiosity, brilliance, and vision required to learn hardware engineering.
But now AI changes the equation. Now everyone can code. The software engineers define the architecture. The hardware people bring the physics and engineering know-how.
We have a turbomachinery design tool that lets one engineer do in realtime what previously took 100 engineers six months. Another tool auto-generates 65% of entire aircraft structure. I’m sure we’ll figure out the other 35% too. Eventually these should not be multiple tools, but one integrated physics model of a complete vehicle, airframe and powertrain.
The results speak for themselves. A boomless supersonic airplane built by 50. A full scale supersonic turbine made by 150. And with this approach, I don’t think we’ll need that many more people to build and certify a supersonic airliner.
You don’t need a big team to build big things—not when everyone on the team is a software engineer.
More than a decade in, the most insane realization is that software is the key to reaccelerating hardware engineering. But in a way that I don’t think is widely understood or implemented, even in the burgeoning hard tech community.
Most hardware engineering is actually software engineering in disguise. It’s spreadsheet “code,” engineering calculations done as one-offs. Eg wing structural loads in, spar sizing design out. Thermal loads in, air conditioning system sizing out.
Everything is a calculation, and calculation is software. But these calculations are done essentially by hand, as if we are doing arithmetic but have not yet discovered algebra.
Hardware engineers have baby software, built in spreadsheets, the kind of crappy software I wrote as a kid before I understood architecture, unit tests, APIs, version control, and everything else that makes modern software good.
Our airplanes and engines are built with the software sophistication of a script kiddie.
So what we tell our engineers: don’t design the hardware, design the *algorithm* for designing the hardware. And take the algorithm seriously—as an algorithm and as a first-class engineered software system.
Never design a piece of hardware, such that you couldn’t redesign it tomorrow with a different set of assumptions.
This all sounds obvious, yet our experience at Boom is that it changes everything.
Ten years ago we couldn’t afford to build much software. We had a small team—and there were few software engineers with the curiosity, brilliance, and vision required to learn hardware engineering.
But now AI changes the equation. Now everyone can code. The software engineers define the architecture. The hardware people bring the physics and engineering know-how.
We have a turbomachinery design tool that lets one engineer do in realtime what previously took 100 engineers six months. Another tool auto-generates 65% of entire aircraft structure. I’m sure we’ll figure out the other 35% too. Eventually these should not be multiple tools, but one integrated physics model of a complete vehicle, airframe and powertrain.
The results speak for themselves. A boomless supersonic airplane built by 50. A full scale supersonic turbine made by 150. And with this approach, I don’t think we’ll need that many more people to build and certify a supersonic airliner.
You don’t need a big team to build big things—not when everyone on the team is a software engineer.
153
340
4,052
407,325
Pedro Oliveira retweeted
If you haven't played with the Devcon 2027 Lisbon website you're missing out.
Seems @nloureiro left a bunch of easter eggs in there :p
Change directory to games and ls your way around or ssh to past devcons! Don't forget to customise settings!
Seems @nloureiro left a bunch of easter eggs in there :p
Change directory to games and ls your way around or ssh to past devcons! Don't forget to customise settings!
Join the waitlist: devcon.org/lisbon/
3
5
46
1,261
If you're coming for Web Summit this is probably the only side-event you need to attend.
luma.com/el6vmqrg
luma.com/el6vmqrg
2
4
13
660
Pedro Oliveira retweeted
Devcon 9 is coming to Lisbon! 🇵🇹
October 11–13, 2027
Building on Mumbai, we're bringing Ethereum's builders together with more friends and allies working toward a freer, more open internet.
October 11–13, 2027
Building on Mumbai, we're bringing Ethereum's builders together with more friends and allies working toward a freer, more open internet.
132
219
1,630
153,169
Pedro Oliveira retweeted
Back in March, our team submitted a formal proposal for #Devcon9 to be hosted in Portugal's capital, Lisbon.
Since then we've been lobbying for this to happen around the clock.
Live from Mumbai, we're so happy to share the official announcement: Devcon 2027 will be in Lisbon!
Since then we've been lobbying for this to happen around the clock.
Live from Mumbai, we're so happy to share the official announcement: Devcon 2027 will be in Lisbon!
The Lisbon Devcon push is real. We're taking the lobby work to Cannes during @EthCC.
Me, @pcbo, @onunoreis, @huvoliveira, @rmdmac and the @onchainlisbon team are hosting a casual get-together to move this forward. First round is on @ScopeLift.
Register here 👇
luma.com/0joyg1nw
29
14
181
7,156
Pedro Oliveira retweeted
As a proud Portuguese, I’m ashamed to see AIMA question James’s qualifications and doing this. He built and leads some of our best engineering teams in Lisbon. His track record is well documented. Portugal should welcome him—not put his and his family’s residency at risk.
just received notice that an Eng leader here at Cloudflare who moved to Portugal from the US has had his residency permit denied
in the months he has been waiting for one, he has hired five full-time engineers and six interns here in Lisbon. alongside the hundreds of others we residencia_pt the immigration insanity here is making it increasingly hard to just operate a tech office in Lisbon much less grow one
13
27
276
18,403
Pedro Oliveira retweeted
Pre-seed traction isn't revenue.
It's shipping speed, who you've hired, customers you shouldn't have yet, and how fast you learn.
David Tisch said it well on 20VC: the great ones get so far ahead you're confused how.
Finding those founders is my sole focus.
It's shipping speed, who you've hired, customers you shouldn't have yet, and how fast you learn.
David Tisch said it well on 20VC: the great ones get so far ahead you're confused how.
Finding those founders is my sole focus.
50
6
248
14,124
Pedro Oliveira retweeted
The most profitable humanoid deployment in the next 2 years will be in the sex industry.
@hedgerati0, ex-CPO at Slush, has done 60+ one-on-ones with the best robotics founders, researchers, and mechanical engineers in SF. Here’s what he learned:
Data
- The hunger for data is at le are hungry for is changing.
- Raw first-person footage mined in China, India and Southeast Asia still sells. It just barely moves the models anymore.
- He heard it again and again: generic egocentric data is saturated, low margin, and the middleman space is crowded.
- Henri’s bet is that raw egocentric data is fully saturated within a year or two. Demand then shifts to data that is labeled, mapped, collected on known equipment and specific robot hardware, plus the failures of robots already in the field.
- Labs are already asking for this and the budgets are real. It is months away, not years.
- Labeling is the quiet bottleneck. Ask ten operators to pick up a spoon and you get ten different ways of doing it. Without labels that describe what happened in between, more demonstrations add confusion, not skill.
- Touch and force data is the next thing labs cannot get from video. A model can know a glass is fragile and still have no idea how hard this particular hand should squeeze.
- He has talked with people building hands and force-sensing rigs. The interesting part: their buyers are the labs, not anyone deploying robots in the field.
- Even the labs building VLA models (the models that turn vision and language into robot actions) don’t know which data improves their models fastest. Everyone is buying blind.
- The honest data sellers already know this and are moving from selling hours to selling tooling, labeling and evals.
Models
- OpenAI’s Astra changed the conversation. A general model, not a robotics model, picked up a block and put it in a bowl 19 times out of 20, and beat specialized robotics models on a large set of simulated manipulation tasks.
- It was probably trained on little or no robot data. It just understands the world well enough to reason its way through the task.
- He sees this as a completely different approach to the edge-case problem.
- The old way: collect every edge case from the real world, one demonstration at a time.
- The new way: a model that already understands physics, plus simulation to run through endless edge cases cheaply.
- In his view, simulation doesn’t finish the last mile of data collection. It replaces most of the collection.
- Astra is nowhere near the nines. Each attempt costs real money and takes long seconds to think. You can’t run a warehouse on that.
- So VLAs win the near term, and they are moving in the right direction. But VLAs are a few steps ahead on a slow road, and general models are on a fast one.
- He expects the general models to catch up within a couple of years, and wouldn’t be surprised if it’s sooner.
- When they do, the brain of the robot is a general model like Astra. What stays separate is the reflex layer: the controller that tightens the grip the instant the glass starts slipping. A reasoning model can’t think fast enough for that.
- That layer stays small, fast and specific to the hardware, and may end up being plain control software rather than a trained model at all.
- General models have taken over verticals one by one. Robotics is one of the next ones.
- Standalone robotics models will have to prove what they can do that a general model can’t, and that list is getting shorter.
Deployment
- It’s very manual. Founders or engineers have to be on site, and the first job is earning the customer’s trust.
- A loose wire or a bad camera angle means someone gets on a plane. One company flew an engineer from California to Spain to swap a memory card.
- The prerequisite for any deployment, not just pick-and-place, is accuracy as close to the nines as possible.
- A robot that works almost every time still fails several times a day, and the failures eat the savings.
- Customers expect near-perfect reliability and better than human speed. Demos are easy, production is hard.
- Because of all this, robotics scales linearly, one site at a time, not exponentially like software.
- What’s possible in two years will be mind-blowing. The world itself won’t look that different.
What surprised him most
- Investors still judge robotics with SaaS logic and ask about ARR, when hardware revenue looks nothing like that early on.
- The data gold rush is a short-term play, not a moat. He was surprised how many VC-backed startups are betting on it growing.
- Deployment itself is becoming the product. VCs now back companies that deploy for the sake of deploying, to collect edge cases and failures.
- More than one startup has pivoted from automating a task to just deploying and selling the data to labs. One of them was building robot hands for a cloud kitchen.
- Outside entertainment, humanoids are a gimmick for the next couple of years. Real progress comes from narrow, hyper-specific robots.
- The home vacuum is still robotics’ best business, and it’s over a decade old.
Where the bottleneck goes next
- Data today. Then models. Then deployment. Then hardware.
- Once hardware is a solved, sold product, we’re living in utopia.
- Hardware means more than actuators and batteries. If the brain is a frontier-size model, it needs more memory and compute than fits on a robot today. The chip on the robot becomes a real constraint.
- China is years ahead on hardware because of supplier density. A prototype takes hours in Shenzhen and weeks or months in the West.
- The US is pushing back with import restrictions and buy-American rules. That will shape where robots get built, not how fast they get smart.
- The most valuable skill he’s found: being wrong fast.
- He changed his view of this space several times in a month. Every time, it was because he asked a dumb question instead of pretending he knew.
- He isn’t an engineer. His way in was open-source models and kits like Hugging Face’s SO-101.
- Now he’s building an underwater robot that cleans cargo ship hulls.
@hedgerati0, ex-CPO at Slush, has done 60+ one-on-ones with the best robotics founders, researchers, and mechanical engineers in SF. Here’s what he learned:
Data
- The hunger for data is at le are hungry for is changing.
- Raw first-person footage mined in China, India and Southeast Asia still sells. It just barely moves the models anymore.
- He heard it again and again: generic egocentric data is saturated, low margin, and the middleman space is crowded.
- Henri’s bet is that raw egocentric data is fully saturated within a year or two. Demand then shifts to data that is labeled, mapped, collected on known equipment and specific robot hardware, plus the failures of robots already in the field.
- Labs are already asking for this and the budgets are real. It is months away, not years.
- Labeling is the quiet bottleneck. Ask ten operators to pick up a spoon and you get ten different ways of doing it. Without labels that describe what happened in between, more demonstrations add confusion, not skill.
- Touch and force data is the next thing labs cannot get from video. A model can know a glass is fragile and still have no idea how hard this particular hand should squeeze.
- He has talked with people building hands and force-sensing rigs. The interesting part: their buyers are the labs, not anyone deploying robots in the field.
- Even the labs building VLA models (the models that turn vision and language into robot actions) don’t know which data improves their models fastest. Everyone is buying blind.
- The honest data sellers already know this and are moving from selling hours to selling tooling, labeling and evals.
Models
- OpenAI’s Astra changed the conversation. A general model, not a robotics model, picked up a block and put it in a bowl 19 times out of 20, and beat specialized robotics models on a large set of simulated manipulation tasks.
- It was probably trained on little or no robot data. It just understands the world well enough to reason its way through the task.
- He sees this as a completely different approach to the edge-case problem.
- The old way: collect every edge case from the real world, one demonstration at a time.
- The new way: a model that already understands physics, plus simulation to run through endless edge cases cheaply.
- In his view, simulation doesn’t finish the last mile of data collection. It replaces most of the collection.
- Astra is nowhere near the nines. Each attempt costs real money and takes long seconds to think. You can’t run a warehouse on that.
- So VLAs win the near term, and they are moving in the right direction. But VLAs are a few steps ahead on a slow road, and general models are on a fast one.
- He expects the general models to catch up within a couple of years, and wouldn’t be surprised if it’s sooner.
- When they do, the brain of the robot is a general model like Astra. What stays separate is the reflex layer: the controller that tightens the grip the instant the glass starts slipping. A reasoning model can’t think fast enough for that.
- That layer stays small, fast and specific to the hardware, and may end up being plain control software rather than a trained model at all.
- General models have taken over verticals one by one. Robotics is one of the next ones.
- Standalone robotics models will have to prove what they can do that a general model can’t, and that list is getting shorter.
Deployment
- It’s very manual. Founders or engineers have to be on site, and the first job is earning the customer’s trust.
- A loose wire or a bad camera angle means someone gets on a plane. One company flew an engineer from California to Spain to swap a memory card.
- The prerequisite for any deployment, not just pick-and-place, is accuracy as close to the nines as possible.
- A robot that works almost every time still fails several times a day, and the failures eat the savings.
- Customers expect near-perfect reliability and better than human speed. Demos are easy, production is hard.
- Because of all this, robotics scales linearly, one site at a time, not exponentially like software.
- What’s possible in two years will be mind-blowing. The world itself won’t look that different.
What surprised him most
- Investors still judge robotics with SaaS logic and ask about ARR, when hardware revenue looks nothing like that early on.
- The data gold rush is a short-term play, not a moat. He was surprised how many VC-backed startups are betting on it growing.
- Deployment itself is becoming the product. VCs now back companies that deploy for the sake of deploying, to collect edge cases and failures.
- More than one startup has pivoted from automating a task to just deploying and selling the data to labs. One of them was building robot hands for a cloud kitchen.
- Outside entertainment, humanoids are a gimmick for the next couple of years. Real progress comes from narrow, hyper-specific robots.
- The home vacuum is still robotics’ best business, and it’s over a decade old.
Where the bottleneck goes next
- Data today. Then models. Then deployment. Then hardware.
- Once hardware is a solved, sold product, we’re living in utopia.
- Hardware means more than actuators and batteries. If the brain is a frontier-size model, it needs more memory and compute than fits on a robot today. The chip on the robot becomes a real constraint.
- China is years ahead on hardware because of supplier density. A prototype takes hours in Shenzhen and weeks or months in the West.
- The US is pushing back with import restrictions and buy-American rules. That will shape where robots get built, not how fast they get smart.
- The most valuable skill he’s found: being wrong fast.
- He changed his view of this space several times in a month. Every time, it was because he asked a dumb question instead of pretending he knew.
- He isn’t an engineer. His way in was open-source models and kits like Hugging Face’s SO-101.
- Now he’s building an underwater robot that cleans cargo ship hulls.
14
3
62
4,330
Hybrid warfare is obvious.
🚨🇵🇱🇩🇰 UPDATE: TWO AIRCRAFT TARGETED BY BOMB THREATS IN POLAND
Further details have emerged following two bomb threats involving passenger aircraft arriving in Warsaw this morning.
🇩🇰 SAS flight SK751
• Copenhagen → Warsaw Chopin
• Airbus A320-251N
• Seats 157 - number of rmed.
• Landed safely at 09:35
• SAS confirms a threat was received.
🇵🇱 LOT Polish Airlines flight LO286
• London Heathrow → Warsaw
• Boeing 737 MAX 8
• 181 people on board
• Diverted to Warsaw-Modlin, landing at 10:01
• Passengers have been evacuated as security checks began.
Warsaw-Modlin Airport was temporarily closed following the incident.
NOTABLY: A separate Wizz Air flight was diverted to Budapest yesterday following another bomb threat.
There is currently NO evidence linking these incidents or suggesting a coordinated operation.
Authorities are investigating.
0
0
5
559
Hybrid warfare is coming for all of us.
BREAKING: Europe is under a massive widespread cyber attack right now Sweden, Belgium, Luxembourg, Estonia, Ireland the Netherlands and just now Finland are all experiencing disruptions to supermarket refrigeration systems causing food to go off.
The attack is spreading...
4
0
11
1,927
Pedro Oliveira retweeted
Robotics is solved when world GDP has grown by a hundred orders of magnitude and we've collonized the galaxy, not when SOTA on some benchmark goes from 42.2% to 63.1%
29
56
592
32,309
Pedro Oliveira retweeted
Karpathy just fixed YouTube in 2 hours. On a weekend.
> He got tired of YouTube's recommendations
> So he had Claude go through 43 million Hacker News comments from the last 20 years
> It found the YouTube videos smart people kept recommending again and again
> That boiled down to 6,454 videos actually worth watching
> He turned them into a site called HackerTube
> Total build time was about 2 hours with Claude Opus 5.5
> No algorithm chasing your clicks, just 20 years of real recommendations
Cool stuff I would say 🫡
> He got tired of YouTube's recommendations
> So he had Claude go through 43 million Hacker News comments from the last 20 years
> It found the YouTube videos smart people kept recommending again and again
> That boiled down to 6,454 videos actually worth watching
> He turned them into a site called HackerTube
> Total build time was about 2 hours with Claude Opus 5.5
> No algorithm chasing your clicks, just 20 years of real recommendations
Cool stuff I would say 🫡
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org
- Excellent (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
23
138
2,464
335,141