Prepare your answers.

1. Your name

2. Email

3. Where you're based

4. Links: site, LinkedIn, X

   Website, LinkedIn, X, whatever shows your work. One line, comma-separated.

1. Startup name

2. Industry

   Pick one: AI / ML · Fintech · Health · Consumer · SaaS / B2B · Climate · Crypto / web3 · Other

3. What are you building? (one line)

   This is the one place a single sentence is exactly right. What you do, for whom, and the outcome; concrete, no buzzwords.

   Solo founder example

   ProbeStack lets engineering teams replay real customer workflows against a new AI-agent version in a safe sandbox before they ship it.

4. Stage

   Pick one: Just an idea · Prototype · Launched · Making revenue · Raised a round

5. Funding to date

   Pick one: Bootstrapped · Pre-seed (<€250k) · Seed (€250k–2M) · €2M+ · Prefer not to say

1. What problem are you solving, and who has it?

   A few sentences: name the pain, who feels it most acutely, and how they cope today. Specific beats broad.

   Solo founder example

   Teams can unit-test an agent's code, but they cannot reliably test what a model or prompt change will do across a real multi-step workflow. A release can pass a small set of canned evals and still retry a payment, update the wrong CRM record or get stuck after an unusual tool response. Today engineers inspect traces by hand, maintain brittle test fixtures and use production canaries to catch what they missed. The teams we interviewed spend one or two engineer-days on every important model upgrade and still do not trust the result.

2. Why you? Your unfair advantage or unique insight

   A short paragraph on your unfair advantage, the insight, access, or ability most people don't have. Earned, not aspirational.

   Solo founder example

   I spent five years building workflow infrastructure at a B2B fintech and owned the incident review after an internal agent retried failed transfers. I then built a replay harness that was adopted by 14 engineering teams before every model release. I know where nondeterministic systems actually fail—tool state, retries and permissions, not just the final text—and former teammates at four agent startups have given me production traces and weekly access to their release process.

3. Who else is doing this, and why will you win?

   A paragraph: who else solves this (even 'a spreadsheet' or 'nobody'), and the specific edge that lets you win.

   Solo founder example

   The status quo is a home-grown eval script plus a dashboard that scores individual outputs. Those tools are useful, but they do not execute the whole workflow against the state the agent saw in production. ProbeStack records the event graph, rebuilds the relevant state in an isolated environment, intercepts outbound writes and compares the resulting state rather than only comparing text. That catches failures such as a duplicated refund even when the response sounds correct. Our risk is integration breadth: if every customer's tool stack needs custom work, the product becomes a services business.

4. Traction so far

   A paragraph with real evidence someone wants this. Numbers help: users, revenue, growth, waitlist, pilots, retention. If you're pre-traction, say what you've learned instead.

   Solo founder example

   Four design partners have shared 18,400 redacted production traces. Two of them used ProbeStack during recent model migrations: we replayed 6,200 workflows and found 31 regressions before release, including three that would have triggered external writes. Both are now paying €1,200 per month; the other two are unpaid design partners. Installation still takes about six engineer-hours and the SDK only supports TypeScript, so usage is not yet self-serve.

5. How big can this get if it works?

   A paragraph: how big this gets if it works, and the ambition behind it. Reason from your wedge out to the whole.

   Solo founder example

   We are starting with software teams whose agents can take actions in payments, support or internal operations, where a silent regression is expensive enough to block deployment. At an expected €12k–€60k annual contract, 5,000 such teams would be a €100m+ recurring business. If ProbeStack becomes the trusted pre-production environment for agents, the larger product is a deployment control plane: replay, approval, gradual rollout and rollback across models and tools. We have evidence for replay; we do not yet have evidence that customers want the wider platform.

6. What will you build during the week?

   A paragraph on the one thing you'd ship in seven uninterrupted days. Concrete and demoable beats grand and vague.

   Solo founder example

   I want a Python team to go from `pip install` to its first useful replay in under 30 minutes. That means capturing one agent run, redacting secrets, replacing its external tools with safe adapters and showing a state diff when the replay finishes. On Friday, an engineer from our third design partner will attempt the setup from the public docs while I watch silently. If I have to edit their fixtures or explain our event model, the build has failed.

We back people, not résumés. Experience isn't a filter here; this is the part we actually read.

1. Why are you really building this?

   A few honest sentences on why you're really building this, not the pitch version. We read this one closely.

   Solo founder example

   I was on call when our agent retried 86 transfers after a timeout and every test we had was green. The technical fix took a day; rebuilding trust with operations took months. I kept thinking about how many teams are now deploying systems that can act but testing them as if they only generate text. ProbeStack is the infrastructure I wanted that night, and I want to make shipping an agent feel as disciplined as shipping any other production system.

2. Tell us about a time you kept going when quitting made more sense.

   Tell the story: what happened, what quitting would have looked like, and what you actually did. Specifics make it real.

   Solo founder example

   The first design partner paused our pilot because replays passed while the same workflows failed in production. I had modelled each tool call correctly but ignored the ordering of asynchronous events between them. For three weeks I rebuilt the runner around an append-only event log, manually reduced 400 failing traces and wrote a deterministic clock for retries and timeouts. We reproduced 37 of the 41 production failures and earned the pilot back. It was slow work caused by an architectural mistake I had made, which is exactly why walking away was tempting.

3. What are you like to work with?

   A short paragraph on what people who've worked with you would actually say, the good and the sharp edges.

   Solo founder example

   Former teammates would say I am calm during incidents and unusually willing to reduce a vague failure to a reproducible test. They would also say I can turn the third occurrence of a problem into a framework before checking whether a small fix would do. I now write down the concrete customer failure before proposing an abstraction and will not add a new primitive until it explains three real traces. Under time pressure, I still need someone to challenge me on that.

4. Something you've changed your mind about recently, and why.

   A paragraph on something you've genuinely updated on lately, whether your market, your product, or yourself, and why.

   Solo founder example

   I started by treating deterministic replay as the goal: same inputs, same tokens, same output. That produced noisy failures whenever the model phrased a correct answer differently and missed cases where different wording caused the same bad side effect. I now think the useful unit is an invariant—no duplicate payment, correct permission boundary, task completed—not an exact transcript. We still offer deterministic fixtures for debugging, but release decisions are based on state and behaviour.

5. Why noclue, why now?

   A paragraph: why this program, and why now for you and the company. Specific to noclue beats generic.

   Solo founder example

   The next constraint is narrow enough for a focused week: make the replay engine useful without me designing every fixture. I can write the Python SDK at home, but I want technical founders to attack its abstractions and a non-customer to attempt the setup from scratch. noclue is attractive because the week ends with a concrete test, not a week of general advice. I would arrive with a partner's redacted traces and leave with either a self-serve integration or a clear reason the architecture is still too bespoke.

6. CV or résumé (optional)

   Genuinely optional. We look for people, not CVs, so only add one if it helps you tell your story.

7. Pitch deck (optional)

   If you have one. A few rough slides is plenty; we don't need a polished raise deck.