The six-week AI sprint: how one VC team actually adopted AI

Ask most firms how AI has changed their work and you get vague answers about efficiency. Ask Melissa Montan of March Capital and the answer is total. Pressed for the single biggest way AI has changed the firm, she doesn't narrow it down. "The one way is everything. I don't know what I do now where I don't use AI."

That kind of saturation doesn't happen by handing everyone a chatbot login and hoping. At March, it happened because the four-person platform team ran a deliberate, time-boxed experiment: a six-week AI sprint, one workflow per person, with weekly check-ins. It's the rare AI-adoption story that comes with a method other teams can copy, which is exactly why it's worth walking through.

The sprint: one workflow per person, six weeks

The structure is almost aggressively simple. As Montan describes it: "We kicked off in our team an AI sprint. We spent six weeks. Each person was owner of one workflow, and week in and week out, we'd have touchpoints to touch base on progress, if we were hitting any blockers, if we needed to pivot."

Here's the structure:

First, one workflow per person: instead of asking everyone to "use AI more" across everything they do, each team member owns a single, concrete process to rebuild. That makes the goal specific and the ownership clear. Second, a fixed six-week window: long enough to make real progress, short enough to create urgency and a definite endpoint. Third, weekly touchpoints focused on progress, blockers, and whether to pivot, so problems surface early and nobody spends five weeks stuck on a dead end in private.

What came out of it

A six-week sprint sounds like the kind of internal initiative that produces a slide deck and little else, but March's produced structural change. Montan lists what actually shifted: "some really substantial changes, whether it's re-platforming because we found a better platform that has better integrations, fixing our infrastructure, and ultimately changing the way we output things that we do repeatedly throughout the year."

Re-platforming means they replaced tools during the sprint once a workflow revealed a better-integrated option. Fixing infrastructure means the sprint surfaced and corrected underlying data and systems problems, not just surface tasks. And changing how they produce recurring work means the gains compound: any process the team repeats through the year now runs differently, permanently.

The stack they landed on

Since each person was solving a real workflow, the tools they ended up with are the ones that survived contact with actual work. Montan describes the progression: "We started with Perplexity and ChatGPT, and more recently rolled out Claude firm-wide, and that has absolutely been a game changer. So we are day in and day out in Claude Cowork." The team also built its own connectors to tie systems together: "We've also learned how to create our own MCPs to connect our various databases."

Madeline Pope adds the content side of the stack: "We use Opus Clip for our short form content. We use one of our portfolio companies, Canva, every single day." (Canva is a March portfolio company, which is its own small lesson in dogfooding what you back.) Ultimately the team assembled a working toolkit through use, kept what earned its place, and connected the pieces rather than treating any single tool as the whole answer.

The pace, and the one rule

It's important to remember how fast AI is moving. "Things we're doing in Claude Cowork weren't possible three months ago," Montan notes. That's the argument for running a sprint now instead of waiting for the tools to settle. They won't settle, and a team that has built the muscle to adopt quickly compounds that advantage every quarter.

The team also set guardrails about AI-generated content. Montan calls it a "time-saver when done well." Pope sharpens it: "Has to be done well, but we've seen it save hours of our team's time." The output still has to meet the bar. Speed is only a win when the work is good.

How to run your own six-week AI sprint

The borrowable version of March's method, stripped to its structure:

1. Pick one workflow per person. Give every team member a single, concrete process to rebuild with AI, not a vague mandate to use it more.
2. Box it to six weeks. Long enough for real progress, short enough to force momentum and a clear finish line.
3. Hold weekly check-ins on three things: progress, blockers, and whether to pivot. Surface dead ends early, in the open.
4. Let the work choose the tools. Start with what you have, and adopt or replace based on what a real workflow actually needs, including building your own connectors between systems.
5. Hold the quality bar. Treat AI output as a draft that still has to meet your standard. The time savings only count if the work is good.
6. Expect structural change, not just speed. The real payoff is re-platformed tools, fixed infrastructure, and recurring work that permanently runs better.