We're betting on the human side of agentic coding.
aiworklab is the only AI engineering tool that treats human skill as a measurable, compounding asset instead of a side-effect to optimise away. Here's what we believe, and what we're trying to build.
The bet
The current generation of AI coding tools has solved an enormous productivity problem. It has also created a new one nobody is paid to solve: the cognitive steps that used to produce durable learning — retrieval, elaboration, explanation — have been quietly removed from the loop. Engineers are shipping more code and remembering less of it.
This isn't a moral failing on the part of any user, and it isn't a bug in the tools. It's what the interface rewards. The accept-tab interaction loop is structurally hostile to the way humans actually learn things. The longer we stay on it without a counterweight, the wider the gap grows between "people who can drive an agent" and "people who can reason about a system."
We think this gap will become the dominant story of engineering hiring and management between 2027 and 2030. We started aiworklab to build the counterweight.
The next phase of AI tooling won't be measured by how much it does for the human, but by how much better the human becomes while it is doing it.
What we believe
A handful of principles that govern every product decision we make.
Friction is a feature — but only where it teaches.
We meter friction with a per-user skill graph. If you've already demonstrated a concept, the product gets out of your way. If you haven't, you get one beat of friction at exactly the right moment. The curve we want is the opposite of every other AI tool: less interruption as you grow, not more.
Teach on the work, not on toys.
LeetCode-style synthetic problems train you for hiring games, not for the work that follows the offer. Every prompt, card, and check in aiworklab is grounded in code from your own repository. Spaced retrieval pulls questions from code you wrote weeks ago.
Local-first. BYO model. Always.
Skill graph and history live on-device by default. Cloud sync is opt-in. Enterprise can self-host the entire system. We never proxy or markup inference traffic — you bring the LLM, we charge for the teaching layer. This is a permanent design choice, not a launch limitation.
Stand on the shoulders of giants.
We don't rebuild the agent loop. We integrate with — and contribute upstream to — T3 Code, OpenCode, Codex CLI, and Claude Code. The interesting product work is in the layer above, where nobody else is paying attention.
Honest measurement over flattering metrics.
"Lines of code accepted" is anti-correlated with our north star. We optimise for concept retention and solo throughput trend — the user-only output during scheduled fly-solo sessions where the agent is read-only. These are harder to game and tell the truth.
Origin story
aiworklab started from a question we kept asking ourselves and nobody had a satisfying answer to: "if our agent vendor went down tomorrow, what would actually happen to our team?"
Most engineering managers we asked didn't know. The honest ones admitted they were watching juniors stall on promotion, mid-levels lose their debugging instincts, and even some seniors begin to drift. Nobody wanted to say it out loud, because saying it out loud sounds like Luddism. But the discourse shifted in 2025 from "is this real?" to "what do we do?", and that's the gap we exist to fill.
We are a small team. We are not pretending we have the entire answer. We are pretending — well, claiming — that the answer is closer to "an instrument that meters friction intelligently and tells you the truth about your skill" than it is to "another agent that closes tickets faster." If we're wrong, we'll have built a useful workspace and learned something. If we're right, the company writes itself.
A small team. Senior operators. All in.
aiworklab is being built by a small founding team with deep dev-tools and engineering-management experience. We're hiring carefully and slowly — most listings are on the careers page.
We're a tight founding group with complementary backgrounds across developer tooling, engineering management, applied ML, and product design. Several of us have worked at or founded companies in the dev-tools space before. We share one conviction: the next decade's AI tooling should be measured by what it does to engineers, not just for them.
We're based between San Francisco and remote, with a preference for people in Pacific or European time zones so we can overlap for a few hours each day.
People who get it.
A growing list of operators, researchers, and writers who are thinking publicly about what AI is doing to the human side of the craft.
Engineering management
"We're heading into a decade where the senior engineer becomes the most underpriced asset in the market."
Cognitive science
"The single most well-replicated finding in learning research is that retrieval beats re-reading. AI tooling has quietly stripped retrieval from the loop."
Industry research
"The companies that pay attention to skill telemetry between now and 2028 will have a structural advantage in the decade after."
If this resonates, come work on it.
We're hiring senior engineers, a founding designer, and a devrel. We want people who care about the craft.