Half the job is coaching, not capital
We are building a quant VC. Most weeks, the work is supposed to be dashboards, scoring pipelines, founder backtests, and conversations with our research team about the next experiment. That part exists. But after a decade of taking founder calls, we noticed something we had not planned for: roughly half of the job had quietly become coaching and psychology.
Founders do not usually call us to talk about cap tables. They call to talk about a co-founder who is checked out, a hire they have to fire, a round they cannot sleep through, a metric that looks fine on the surface and stuck underneath. Most VCs do not say this part out loud. We had been doing it on the phone for years, one founder at a time, and there was no obvious way to do more of it without adding partners.
Vela Coach is the open-source AI coach for founders we built to productize that work. It reads your last 90 days of meetings (investor conversations, team one-on-ones, founder calls, sales calls) and surfaces the patterns a $500-an-hour executive coach would surface after a month of listening. Free, browser-only, self-hostable.
Why founders need an AI coach before pitching VCs
The first place this earns its keep is right before a VC pitch. A VC partner sitting across from a founder in a 30-minute first meeting is doing a kind of pattern matching that founders rarely get to see from the outside. Within the first ten minutes, an experienced investor has already formed a view on how the founder thinks under pressure, how they handle uncertainty, where their real conviction is, and where they are performing.
Most pitch coaching is about the deck. The deck is the easy part. The hard part is everything you have already said in the past 90 days, because that is what shaped how you talk now. The phrasing you use with your team about the roadmap leaks into how you describe traction. The way you handle a tough one-on-one shapes how you handle a tough investor question. The number you fudged in last month's board update is the number you will fudge again on Tuesday at 10am.
The most useful pre-pitch preparation is not rehearsing answers. It is a structured audit of how you have actually been talking, across the conversations you have had over the past three months. Most founders never get that audit. The ones who do usually pay an executive coach thousands of dollars a month for it. An AI coach for founders compresses that audit into minutes, at zero marginal cost.
What an AI coach for founders actually reads
The transcripts go in raw. The report comes out structured. Vela Coach reads anything recurring that shapes how you run the company (investor conversations, team one-on-ones, founder peer calls, sales calls, board updates) and produces a coaching report across the dimensions a human coach would track:
- Personality and leadership read. The persistent traits that show up across contexts. How you negotiate. How you delegate. How you signal confidence. How you handle disagreement.
- Hedging and conviction asymmetries. Places where the language you use is more cautious than the underlying conviction warrants, or vice versa. This was the pattern that surprised us most when we ran the tool on our own meetings.
- Cross-context contradictions. Where what you tell investors does not match what you tell your team, or where what you tell your team does not match what you tell yourself.
- Fundraising posture. How you frame traction, what numbers you state with full confidence, what you fudge, what you avoid.
- Team dynamics. Who you escalate with, who you avoid, what topics keep coming back without resolution.
- Recurring stuck points. The things you raise across multiple conversations and never close out.
The output is not a score. It is a structured read, the kind a human coach would write up after a month of sitting in on your meetings, and it takes minutes to produce.
What surfaced when we ran it on ourselves
We ran the tool on our own meetings before shipping it to anyone else. The personality reads were mostly confirmation. High-empathy quant. ENTJ. Most of that we already knew.
The surprise was a consistent hedging pattern. In conversations where the underlying confidence was high, the language was the language of doubt. “I think we can probably hit X” instead of “We will hit X.” From the inside, it felt like politeness. From the outside, it reads as low conviction. For a founder pitching VCs, that asymmetry is a fundraising tax. For anyone running a team, it is a leadership tax.
A good human coach would have surfaced this after weeks of sitting in on conversations. The AI surfaced it from raw transcripts in minutes. That asymmetry is the whole reason we shipped this as a product.
Vela Coach vs traditional executive coaching
The natural next question is how this stacks up against the category it most resembles: executive coaching. The honest answer is that the comparison breaks across four dimensions, and they cut in different directions.
- Cost to first useful pattern. Traditional executive coach: $5,000 to $15,000 and four to six weeks before the first useful named pattern. AI coach for founders: under a dollar of model spend and minutes from upload to report.
- Coverage. A human coach sits in on a small sample of your meetings. Vela Coach reads everything you upload across 90 days, which is usually an order of magnitude more material.
- Quality of the conversation about the patterns. A human coach is still better at the follow-up: deciding what to do with the pattern, which one to confront first, how to bring it to a co-founder. AI is faster at finding patterns; humans are better at the conversation that follows.
- Privacy posture. Traditional coaches are bound by professional confidentiality but still hear everything in real time. Vela Coach runs in your browser, never sends transcripts to a vendor backend, and can be wiped with /reset.
The right model is augmentation, not replacement. Use the AI coach to find the patterns. Take the most important two or three to a human coach, a co-founder, or a trusted investor for the conversation that follows.
How to use an AI coach with Claude Code: end of sprint, before VC outreach
So when does an AI coach actually fit in a founder's week? In our experience, the cleanest moment is the seam between two distinct modes most founders already alternate between. Build mode, usually anchored around Claude Code sessions, where the product gets shipped. And business mode, anchored around investor outreach, sales conversations, and team management, where the company gets built. The two modes use very different parts of the brain. Switching between them is where most founders lose energy.
Vela Coach lives in that seam. The recommended ritual is simple:
- Finish the Claude Code session. Get the build loop closed. Code merged, tests green, deploy clean.
- Run Vela Coach on the last 90 days of meetings. Take ten minutes to read the report.
- Then do the VC outreach. Or the sales call. Or the hard one-on-one. The audit is fresh in your head when it matters.
For developer-leaning founders, the framing that lands best is treating the coach output the way you would treat a code review. A second pair of eyes before you ship. The cost of running it is effectively zero. The cost of going into a fundraise round with an unaudited posture is paid in dilution.
If you also use Cairn (our open-source end-to-end test agent for AI-generated code), the natural sequence is Claude Code, then Cairn for the build, then Vela Coach for the conversations. Build verified, then founder verified.
Why our AI coach is open-source and browser-only
None of this works if founders are not willing to upload their meetings in the first place. We knew going in that they would not be, unless the tool was built around the right constraints. Meeting transcripts are some of the most sensitive data a founder generates. They contain hiring decisions, salary numbers, investor feedback, customer complaints, and what you actually believe about your co-founders. Handing that data to a SaaS coach for centralized storage was a non-starter for us. We made two decisions on day one:
- Open source. Every line of code that touches your transcripts is on GitHub. You can read it. You can fork it. You can host it yourself if you want full control over where the runtime lives.
- Browser only. If you use the hosted version at coach.vela.partners, your transcripts and your AI key stay in your browser. We do not run a database on our side. The /reset command wipes the device. If you close the tab, the state goes with it.
Both decisions cost us product surface area. They make some features harder to build. We are fine with that. The point of an AI coach for founders is to give founders leverage without asking them to trust a vendor with the most sensitive part of their working life.
How a quant VC ended up building an AI coach for founders
Which raises the obvious question: why is a venture firm building this at all? Vela is a quant VC: a fund that uses data, AI, and reproducible research to source, evaluate, and support founders. We have written a lot about how that works on the investing side. The coaching side is a less obvious extension of the same posture.
A quant approach treats the things that are usually called soft skills (leadership, communication, conviction, founder psychology) as patterns in data. Not because those patterns are simple, but because they are real, and because there are signals you can measure if you have access to the right substrate. Meeting transcripts are that substrate. They contain enough signal, when read carefully, to reproduce a non-trivial portion of what a human coach picks up across months of listening.
Vela Coach is the first product we have shipped that turns that view into something founders can use directly. It sits inside a broader family of founder-facing tools we are building, alongside Entrepreneur OS for evaluation and Cairn for engineering verification. All free. Most open source. The bet is straightforward: the more leverage we give founders, the better the founders that show up at our door.
How to start: hosted or self-host
Two ways in:
- Hosted version (free): coach.vela.partners. Bring an AI key, paste in transcripts, get a report. Nothing leaves your browser.
- Self-host: github.com/Vela-Engineering/coach. Clone, run locally, audit the code, modify the prompts, fork the whole tool. PRs welcome.
One piece of advice we keep giving the founders we share this with: run it on your own meetings first, before you suggest it to anyone on your team. The patterns it surfaces are not always flattering. The hedging pattern we found in our own transcripts was not something we would have wanted to read for the first time five minutes before a board meeting. Surface it on your own time, sit with it, decide what to do about it. Then go pitch.
For the use cases beyond pitch prep (sales calls, interview rounds, weekly one-on-ones, the recommended weekly cadence), see our follow-up: How Founders Use an AI Coach: Sales, Hiring, and Leadership.