human in the wire
the one-pager
You can now build your product alone, faster and cheaper. One person with domain knowledge and an AI agent can now build software without the need for a dedicated team.
Hiring engineers or outsourcing technical work to agencies became optional, but it also severed the vital feedback loop between the domain expert and the technical professional. If you build alone, you lose the second pair of eyes that has seen the subtle technical mistakes a hundred times.
What you ship is code, and you still carry full responsibility for understanding it.
You gained speed and lost the experienced person in the room. We think you should keep the speed, and this page is about getting that person back.
The three debts
Technical debtIntent debtCognitive debt
Three kinds of trouble pile up when you build fast, and none of them show up if you don't know where to look. They come from recent software-engineering research on what changes when AI writes the code.
Technical debt lives in code. It's the shortcut that works today and breaks six months from now, right when you've built everything else on top of it. Like financial debt, it accrues interest: the longer it sits, the more it costs to change. This is the debt the industry already knows how to manage, and the one AI is getting good at paying down through refactoring, tests, and review.
Intent debt lives in artifacts, or in the lack of them. It's the reasons behind how your product works: the goals and constraints never written down anywhere you or your AI can look them up. The code runs and the tests pass, yet the product slowly drifts from what you actually set out to build, because nothing outside your head records what that was. Intent is best captured the moment a decision is made: sought later, it is often gone for good, and an AI agent with no record of why will confidently optimise for the wrong thing.
Cognitive debt lives in people. When AI writes the code, you can accept it without ever building the mental model you'd have formed writing it yourself. It all still runs, but you've lost the thread of how it works, so when something needs changing or fixing you can't reason about your own product, and your AI can only follow your lead. This inflates your confidence even when the AI is wrong, so you don't notice the understanding is missing until the moment you need it.
Put simply: technical debt makes your product harder to change, cognitive debt makes it harder to understand, and intent debt makes it harder to know what your product is even for.
Source: Margaret-Anne Storey, From Technical Debt to Cognitive and Intent Debt: Rethinking Software Health in the Age of AI (arXiv:2603.22106, 2026). Storey's model describes shared understanding across a team; when you're building solo with AI, that team is you and your agent, and the debts land the same way.
What we believe
Speed is worth keeping. The fix for building fast has never been building slower. The pace AI gave you is yours to spend, and someone experienced should watch the debts while you do.
Advice beats intervention. Plenty of services will take your project and fix it for you. That makes you dependent and makes them slow, and it's how a quick question turns into a four-week engagement with a six-figure invoice. We read and learn your product, then tell you what we'd do and why, and the actual work goes back to your AI, where it's fast and cheap. Every line stays yours.
Your conversations reveal more than your code. What you say to your AI records what you're trying to build and how well you understand it, which is exactly where intent debt and cognitive debt live. So the wire carries your sessions only: your source code stays on your machine. We keep the transcripts, because watching the debts build over time is what sharpens the advice; we delete them whenever you ask, and a DPA is available on request.
Help has to arrive at your pace. You never wait on us. We read as you build and reach out when something is about to bite, with a specific, plain-language heads-up you're free to ignore. That is why it's a human in the wire rather than a human in the loop: the advice is asynchronous on purpose, so it lands at the speed you're actually working at.
Support queues hand you to whoever picks up next. With us you're matched with an engineer who learns your product and stays with it for the life of your project.
Nobody catches everything, and we won't promise to. For a full security audit or a line-by-line review you'll still want a specialist. We're the experienced second pair of eyes you've been missing.
So we built the wire
YouClaude CodeThe wireYour expertYou
- You build, talking to Claude Code, exactly the way you do now. Nothing about your workflow changes.
- The wire is a small plugin that forwards your sessions. We read every one, without being asked; your repository never leaves your machine.
- Your expert reads your sessions as they come in, watches all three debts, and reaches out when something's about to bite. You decide whether to act on it, and hand the work back to your AI.
It comes back around to you, still in control, now with someone experienced on the wire.
Why we built it
We've spent our careers being the experienced person in the room for other people's builds, and we watched AI make that role optional. The builders around us didn't slow down, and they shouldn't have to: they just kept hitting the kind of trouble a five-minute conversation would have prevented. Human in the Wire is that conversation, kept running for the whole life of your product.
Marek Chotoborski. Product engineer with almost a decade of experience at every scale: from multinational consultancies to indie biotech startups.
Kamil Beker. Eight years of experience at companies like Samsung, BMW, and Huawei, and at a few startups, working on natural language processing, constraint programming, automotive software, and embedded systems.
Get a human on the wire
We're onboarding a small number of founding customers.
We use your email only to reach out about early access. See our privacy policy.
Prefer email? Reach out at
hello@humaninthewire.ai