Devin for Robotics Testing Founded by @theOwenBales & Max Kern backed by @alliance
Joined May 2026
- Tweets12
- Following8
- Followers29
- Likes95
Zynoptes retweeted
Why do robotics companies need @Zynoptes? Because failing is expensive.
As a startup, your job is to fail fast, learn, and iterate. But in robotics, every failure can mean broken hardware, weeks of engineering work, and lost customers. Software startups can afford to fail fast in the field. Robotics startups can't.
We're building Zynoptes to change that.
Break your robot before the world does.
Introducing @Zynoptes (ALL18), an Autonomous Simulation Engineer for Robotics that finds and fixes issues before robots reach the customer's site.
You bring your robot’s task and code. Zynoptes builds the simulations, runs hundreds of scenarios, and validates code fixes so your team can confidently deploy.
Congratulations on the launch, @theOwenBales and @KernMaxim87213!
Is your team busy changing the physical world with your robot? Get in touch with us to break it before the world does.
Break your robot before the world does.
Introducing @Zynoptes (ALL18), an Autonomous Simulation Engineer for Robotics that finds and fixes issues before robots reach the customer's site.
You bring your robot’s task and code. Zynoptes builds the simulations, runs hundreds of scenarios, and validates code fixes so your team can confidently deploy.
Congratulations on the launch, @theOwenBales and @KernMaxim87213!
Zynoptes retweeted
This robot passes its first patrol. Then we change the conditions.
In this preview, @Zynoptes tests 488 variations, showcases 4 failures, and fixes and validates 3. The fourth requires a facilities tweak.
I want this loop to be fast and routine for all robotics teams.
Zynoptes retweeted
A robot's job description is starting to change faster than it's hardware.
Same arm, same gripper, new tasks as the models improve.
If that continues to hold, buying a robot starts to become a bet on what it could learn to do, not just what it does today. E.g. buying $200 claude plan in 2025.
One of the biggest surprises in AI over the last few years has been how well coding agents generalize beyond software.
In a recent essay, MIT professor Philip Isola argued that we may be entering the era of robot-use agents: general-purpose models that can control different robots, write policies, and learn new physical tasks with little or no robot-specific training.
In this episode of Decoded, we're joined by the founders of @theWaddleLabs and @Robocurve, two of the startups whose work helped drive this realization. They're working at the frontier of using general-purpose models to control robots, and their recent demos helped inspire the growing conversation around robot-use agents.
Together, we dig into the research behind that idea, from code-as-policies and vision-language-action models to the harnesses and evals needed to make these systems work in the real world.
00:00 — What Are Robot-Use Agents?
02:11 — From Vision-Language-Action Models to General LLMs
04:07 — The Bitter Lesson for Robotics
07:04 — How Coding Agents Learned to Control Robots
10:41 — How Robots Learn From Experience
14:22 — The Harness as a Form of Robot Intelligence
16:14 — Watching Astra Control a Robot
20:55 — Why General Models May Win in Robotics
26:05 — How Close Are General-Purpose Robots?
If you’re a robotics team that is spending significant time testing, simulating, reproducing failures, and validating fixes aiming to deploy to the field ASAP reach out to us.
We close the idea —> confidently deploying to the field gap.
zynoptes.com
The real unlock here might be that AI doesn’t just simulate the world, but decides what worlds are worth simulating.
Understand the robot -> think of ways it could fail -> generate those scenarios -> test them -> fix -> repeat.
That’s exactly what we’re building at @Zynoptes.
Zynoptes retweeted
The real unlock here might be that AI doesn’t just simulate the world, but decides what worlds are worth simulating.
Understand the robot -> think of ways it could fail -> generate those scenarios -> test them -> fix -> repeat.
That’s exactly what we’re building at @Zynoptes.
Zynoptes retweeted
Getting a robot to work once is becoming easier than proving it works reliably. Zero-shot sim-to-real is real progress. But one successful transfer leaves a lot of conditions untested.
I think the next bottleneck is how quickly teams can find and fix those failures in simulation, before customers find them in deployment. (and why we're building the solution to just that at @Zynoptes!)
We made Autoware's planner 171× faster.
It reset its whole 180M-node search graph before every plan. It had touched 1,000.
497 ms -> 2.9 ms. Output identical.
One of seven public PRs we've sent to Autoware and GTSAM -> zynoptes.com/#findings
Zynoptes retweeted
The next big performance gains in robotics won't come from better hardware. It will come from making much better use of the hardware we already have.
A robot can work perfectly and still have implementation choices buried in its perception, planning, or control stack that are quietly costing latency, compute, or memory. And as robotics codebases get larger and more complex, finding those optimizations manually gets extremely difficult and time consuming.
What's especially interesting is that many of these optimizations can happen at the software level, rather than being tied to squeezing more performance out of one specific GPU or CPU.
That's what we're working on with @Zynoptes: automatically finding those optimizations across robotics codebases, regardless of the hardware they run on.
If you're building robots, drones, or autonomous systems, I'd love to run it on your codebase and show you what your engineers are missing.
Zynoptes reads your entire codebase at once, finds the performance optimizations your tools miss, and implements them. The expensive stuff almost never sits in one file. It sits between files, in the call paths, where linters and code review just don't catch it.
AI is writing more code than ever, and optimizing it has become the job nobody has time for. That's the gap we're going after.
We think robotics and AI inference teams have the most to gain here. When milliseconds decide whether your robot works, or GPU costs decide whether your margins work, untapped performance in your repo is money on the table.
Zynoptes finds it in minutes, and every optimization comes with the evidence for why it's real. No guessing.
If you're building in either space we want to run it on your code. Book a demo or DM us: zynoptes.com