Finding Product-Market Fit: How AI Helped Us

Product-market fit is talked about like it's a single moment — the day everything changes. For most companies, it's more like a gradient: gradually more things are working, gradually fewer things feel like you're pushing a boulder uphill.

We're somewhere on that gradient with DenchClaw. Here's what we've learned about finding fit, what AI helped with, and what it categorically cannot do.

The Ironclaw → DenchClaw Pivot#

We started as Ironclaw — a narrower framing of what we were building. The product was essentially the same, but the positioning was "open-source CRM framework" rather than "AI workspace on your Mac."

The Garry Tan tweet and the Show HN response changed something. The comments on the HN post weren't about CRM features. They were about the concept of an AI agent running locally on your Mac with access to all your tools. That's what excited people.

We'd built CRM first because it was concrete. The market was telling us the exciting part was the AI workspace layer that the CRM sits inside.

Renaming to DenchClaw and pivoting the positioning happened within a week of that signal. The product barely changed. The story changed significantly. The response changed immediately.

That's one thing AI helped with: monitoring the signal in real time. We had a DenchClaw instance tracking HN comments, Twitter mentions, and our earliest user conversations. The agent surfaced the pattern — "AI workspace on Mac" was outperforming "open-source CRM" in terms of excitement signals — before we'd consciously noticed it.

Using Your Own CRM to Find PMF#

There's a meta-advantage to DenchClaw for founders: you're using the product you're building to track the data that tells you whether the product is working.

After the Show HN launch, we added every user conversation to our CRM with detailed notes. After 30 conversations, we started querying patterns:

• "What are the most common use cases mentioned in user interviews?"
• "Which features are mentioned in positive vs. negative feedback?"
• "Which user segments show the highest enthusiasm scores?"

The agent synthesized these. The pattern was clear: technical founders and developers building AI tools had the strongest response. They immediately understood what DenchClaw was, could extend it themselves, and had the highest willingness to deal with early roughness.

That's not a surprising finding in retrospect. But getting it from real data in three days rather than three months of intuition is the AI advantage.

The Signals That Actually Matter#

Sean Ellis's "40% test" — would 40% of your users be very disappointed if your product went away — is often cited as the threshold for PMF. We ran this informally by asking users: "If DenchClaw disappeared tomorrow, would you be frustrated, mildly annoyed, or fine?"

The answers were more useful than the percentages. The users who said "frustrated" described DenchClaw as part of their daily workflow. The users who said "mildly annoyed" thought it was useful but replaceable.

The difference wasn't which features they used. It was how deeply it was embedded in their workflow. The users who had connected DenchClaw to their messaging channels (Telegram, Discord) and were interacting with it multiple times per day were the frustrated ones.

That told us where to focus: getting users to deep integration faster. Everything else is noise until users are talking to their CRM on Telegram every day.

What AI Can't Tell You About Fit#

AI can synthesize signals in your data. It can't tell you what signals to collect.

The trap is thinking that because you have AI, you can be less intentional about user research. The opposite is true: you need to be more intentional, because you can now process much more data. The bottleneck isn't synthesis — it's getting the right raw signals into the system in the first place.

No amount of AI can tell you why someone is slightly dissatisfied with your product if they didn't say it explicitly in a conversation. User research requires human curiosity and follow-up. "Interesting, tell me more" is not something you automate.

The Pattern We Found#

After all the data collection and synthesis, our clearest PMF signal is simple:

DenchClaw works best for people who are willing to spend 30 minutes setting it up because they've been burned by cloud SaaS with their data before. These are technical founders, privacy-conscious power users, and people who've had the experience of migrating off a SaaS tool and losing their data.

That's a real segment with real pain, and we serve it genuinely. The question is whether it's large enough and whether it expands over time. I believe it does — as AI agents become more prevalent, the question of "whose cloud are my most sensitive data on?" becomes more pressing for everyone.

Frequently Asked Questions#

How do you define PMF for an open-source product?#

For open-source, I focus on: star growth rate, contributor activity, and how often users recommend it unprompted. Revenue is also a signal if you have it. The traditional SaaS PMF signals apply, but you also have engagement signals from the open-source community.

How long does it typically take to find PMF?#

It varies enormously. Some products find fit in months; most take 1-3 years. The YC average is misleading because YC selects for ideas that have early signal. Plan for two years, be happy if it's faster.

What should you do if you're not finding PMF?#

Talk to your users who stopped using the product. They'll tell you more than the ones still using it. If no one is churning because no one is sticking around long enough, that's a different problem — your onboarding is broken.

Is it possible to have PMF for a product nobody's heard of?#

Yes. PMF is about whether your product solves a real problem for a specific segment — it's not about awareness. You can have deep fit with 100 users and no fit with 100,000. Build depth before breadth.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →