🔥How to Build an AI Harness for Your Investment Firm, One Workflow at a Time

Most funds have access to the same AI as their competitors and a very different amount to show for it. The gap isn't the model. It's everything the firm has built around it.

Every investment firm has accumulated years of knowledge, judgment, rules, and ways of working. But most of that context never reaches the model. It sits in people's heads, scattered documents, and individual chat histories. So every new conversation starts largely from zero.

An AI harness fixes that. It turns what the firm knows and how it works into something AI can consistently use across tasks and team members.

You don't build that harness all at once. You build it workflow by workflow: codifying the knowledge, rules, and judgment behind each task until AI can execute it consistently across the firm.

Here's how to do that in five steps.

Start with one task the team already does regularly.

Screening inbound decks. Drafting the weekly portfolio update. Putting together first-pass LP commentary. Pick one.

Before touching a prompt, map out what that task consists of:

This step is mostly diagnostic. You're mapping the task's real anatomy before deciding what the AI gets fed.

There's no single file that works for every task. What matters here depends entirely on which one you picked in step one.

Deck screening lives and dies on the firm's current investment thesis, the internal version, sharp enough to judge fit against a real deck. A portfolio update runs on KPI history and board notes. LP commentary runs on real fund performance and the tone of past letters.

Figure out which category the task falls into, then write down only what that task actually needs.

If it lands in the thesis camp, this is worth doing properly. The LP-facing thesis is written to persuade, updated once a year if that, full of language that reads well in a deck without doing much operational work. The version the model needs is shorter and current: what's actually getting a deal past the first call this quarter, which of last year's "we invest in X" claims still holds, and which sectors quietly stopped being a yes six months ago even though nobody updated the site.

Whatever gets written down, review it on an actual cadence. A page that's accurate this quarter beats a document from eighteen months ago every time the model uses it to make a call.

Every task like this touches something sensitive eventually, and most firms have never written down where that boundary sits. Do it now, specifically for this task.

Split what you write into two kinds of rules:

Write both down. If the task touches anything MNPI-adjacent or LP-facing, get compliance looking at this list before it goes any further.

What you've just drafted is the seed of a skill for this task. Before writing one from nothing, worth checking whether it already exists. We’ve been curating VC-specific skills with our partner OverDrive at vcskills.com, and a decent version of this task's rules may already be sitting there.

This is the step almost every firm skips, and it's the one that catches whether the last two steps were any good.

Hand the workflow to a different associate. A different team. Someone who wasn't in the room when the rules got written.

Run it against real, recent examples where the firm already knows what happened. Watch closely for two things:

This is where you find out whether the harness generalizes past the one person who built it. That's the entire point of building it in the first place.

Once it's held up through testing, it needs a real decision before it becomes how the firm does this task: leadership has to sign off on it.

This is worth being direct about, because it's usually where things quietly fail. Signing off can't be a partner glancing at a demo and nodding along. It has to mean someone is now accountable for this workflow the same way they'd be accountable for anything else that touches the firm's process.

That raises the question: who owns the tech stack this thing is going to live in?

We covered this exact question in the DDVC Landscape Report 2026. One pattern stood out: the more technical the functional owner of the stack, the higher the AI adoption across the firm.

If nobody at leadership level is willing to own it, the workflow stays a good pilot, however well it did in testing. Once a leader has genuinely taken that on, the workflow becomes the firm's default way of doing this task.

Then the same five steps run again on the next one. A harness gets built this way, one tested, owned workflow at a time.

The five steps above are really one discipline: stop letting AI adoption stall inside one person's chat history. Task by task, write down what the firm knows, what it won't risk, and who's actually accountable for the result.

The edge this year comes down to something smaller than the model itself: whether anyone gave it somewhere to remember what the firm has already worked out.

Pick one task this week. Build the harness around it properly. Everything else follows, one workflow at a time.

Stay driven,
Andre

PS: Join the DDVC Investor Summit at Bits & Pretzels virtually or physically in Munich during Oktoberfest Sep 28-30