Frontier models can write your code and pass your exams. Ask them to design an op-amp and they flail. Atrisa, a Y Combinator P26 startup, is building agents for the last hand-crafted corner of chip design.
There is a strange asymmetry in artificial intelligence right now. A language model can draft a legal brief, refactor a codebase, and explain quantum mechanics to a ten-year-old. Hand that same model the schematic for a two-stage amplifier and ask it to fix the phase margin, and it starts guessing. Analog circuit design - the discipline behind every power regulator, radio, and sensor front-end on the planet - remains stubbornly resistant to the tools that automated almost everything around it. Atrisa is a bet that this is the corner worth taking.
Atrisa, part of Y Combinator's Spring 2026 (P26) batch and based in San Francisco, builds AI agents for analog circuit design. The pitch is compact: describe a circuit in plain English, and the agent helps design it, debug it, and run the simulations. The company frames its product as "an assistant engineer right by your side" - not a replacement for the human, but a tireless second set of hands that has, in Atrisa's words, "internalized the intuition of an analog engineer."
Digital chip design was largely automated decades ago. Logic synthesis, place-and-route, formal verification - the digital world runs on tooling that turns intent into silicon with minimal hand-holding. Analog never got that gift. Designing an industry-grade analog circuit leans on what engineers politely call "tribal knowledge": undocumented rules of thumb, hard-won intuition, and proprietary company data that lives inside senior designers' heads and internal wikis. None of it made it into the training set of a frontier model.
That is the gap Atrisa is built around. A general model has read the internet but has never watched a seasoned designer stare at a Bode plot and mutter that the compensation cap is too small. It doesn't feel parasitics. It doesn't know which topology a given company reaches for when the spec sheet says low noise and low power at the same time. Atrisa's argument is that this knowledge can be encoded - and that whoever encodes it first owns a real moat.
Atrisa's interface is a chat window, but the machinery behind it is specific to circuits. The agent reasons about a design hierarchically - breaking a system into blocks, matching topologies to specifications, and checking its own work against the physical realities that trip up naive designs: reusability, interference, parasitics, and layout. It plugs into the tools engineers already use, including process design kits (PDKs) and existing design documentation, so it augments a workflow rather than demanding a new one.
There is also a multi-agent angle. In a real design team, work stalls when everyone needs the one person who understands the bias network. Atrisa lets engineers hand pieces of a design to collaborating agents so blocks can move forward in parallel, without waiting on a single expert's calendar.
Atrisa's credibility starts with its team. This is not a group of generalists who discovered semiconductors on a whiteboard. The founders worked inside the industry before trying to automate it.
The company was originally named Refortif AI - a detail that still shows up in the founders' email addresses on the refortif.ai domain - before rebranding to Atrisa and sharpening its focus onto analog. Y Combinator group partner Nicolas Dessaigne backs the company.
The honest competitor is the status quo: an experienced human doing the work by hand inside legacy EDA tools from Cadence and Synopsys. General-purpose LLM assistants are the other pole - fast and fluent, but unreliable the moment a problem needs real circuit understanding. A handful of AI-for-chip startups are circling the broader space. Atrisa's distinction is narrowness on purpose: it is built for analog, by analog people.
| Approach | Understands analog | Fits existing workflow |
|---|---|---|
| General LLM assistant | Unreliable | Yes |
| Legacy EDA tools | Via the human | Yes |
| Manual expert design | Deeply | Slow / scarce |
| Atrisa | By design | PDKs & docs |
Analog and mixed-signal design is a choke point in nearly every chip. Every phone, electric vehicle, hearing aid, and satellite depends on analog blocks designed by a comparatively small pool of experts - a pool that isn't growing fast enough to meet demand. If Atrisa's agents let one engineer do the work of several, or let a strong digital engineer venture into analog with a competent guide, the leverage is obvious.
Atrisa's roadmap reads like a company that knows the terrain. Circuit reasoning and design first. Layout next. Post-silicon debug after that. Each step goes deeper into the physical stack, and each is a place where tribal knowledge, not textbook formulas, decides who wins. As a business, the shape is enterprise SaaS: agents sold to semiconductor and hardware teams, wired into their proprietary PDKs and workflows. The company is small - a founding team of ex-semiconductor engineers - and early. But the problem it picked is the kind that rewards depth over breadth.
Whether Atrisa becomes the default copilot for analog design or a sharp acquisition for an incumbent, it is asking a question the rest of the AI industry has mostly skipped: what happens when you point these tools at the work that was supposed to be too hard, too tacit, too human to automate. The answer, for analog at least, is being written now.