This year your fund will buy roughly the same AI as the fund that beat you on your last term sheet.

Same frontier model, same sourcing platforms, same enrichment layers, same chat box. The capability you are paying a premium for is quietly becoming the capability everyone has.

The thing that actually separates the best firm in your category from the median one is not anymore inside any of those tools.

Yes, not having access to these tools is negative alpha. It’s a disadvantage. But for those firms that have access to the same cutting-edge tools, the alpha is somewhere else.

It is the way your sharpest partner reads a data room, the question they ask on the third reference call, the reason they pass on a deal that looks perfect on paper.

That instinct is the asset your firm really sells. And almost no firm has written it down in a form a machine can use.

Every fund does things in ways that are more specific than the partners realize.

How you decide a deal is worth a first call. How you pressure test a market before you believe the TAM slide. What turns a friendly reference into a real one. When a portfolio company moves from "watch" to "intervene."

Some of this lives in your IC template. Most of it is scattered across old memos, Slack threads, deal notes, and the heads of the partners who have seen enough cycles to know better.

That knowledge got treated as background but in reality is a key asset. It’s the taste of your firm.

An agent is only useful when it understands more than the task. It has to understand the method behind the task.

Most firms begin their AI effort with access, and it feels like real progress.

Connect Affinity, wire in the data providers, point the model at the drive, expose the deal notes. That work matters, because a model without access is just guessing.

But access does not produce good judgment. It produces a confident memo that misses the one thing your best partner would have caught on the first read.

A model can ingest every note in the CRM and still not understand how your firm decides to pass.

It can summarize a hundred diligence calls and still miss the signal that tells you a market is about to roll over.

The hard part is not giving the agent more to read. It is teaching the agent how your firm thinks.

A prompt tells the model what to do once.

A skill captures how your firm does a kind of work every time it comes up.

It is the qualification lens a partner applies before they will take a meeting. It is the diligence checklist that exposes a coached reference, and the memo structure that forces the bear case onto the page whether the deal lead wants it there or not.

A skill packages procedure and judgment together. The steps, the edge cases, the questions that always get asked, and the quality bar the partnership refuses to drop below.

The skill is your method, made reusable.

This is not new in software. It is only new in what software can now capture, codify, and make repeatable.

Libraries made code reusable. APIs made services reusable. Workflows made business processes reusable.

Skills make judgment repeatable.

What changed is the executor. For decades a playbook sat in a doc waiting for a human to read it and apply it. At Earlybird, we have an internal wiki that contains majority of our process, assessment, deep dive, post-mortem, and other knowledge.

Now an agent can load the playbook, pull the data, run the analysis, inspect the files, and keep going.

With skills, the playbook stops being a document and becomes a worker.

Picture two funds running the identical frontier model.

The first connected it to their systems. The second connected it to their systems and handed it a library built from the partnership's best work.

The second fund is a different firm. Its agents know how the partnership sources, screens, diligences, writes the memo, supports founders, and updates LPs, not perfectly, but consistently enough to compound.

This is exactly the split showing up in the data for our upcoming DDVC Landscape.

The funds that have operationalized AI into their actual workflows report full scale adoption at roughly 50%. The funds still treating AI as a chat window bolted onto the old process sit at roughly 3%.

Same models available to both. Wildly different firms.

Public skill marketplaces are coming, and most of what they contain is generic. Not too relevant for investment firms. Too much noise, too little signal.

In our DDVC Slack group, we started collecting and curating VC-related skills, and made it available with our partner OverDrive here: https://www.vcskills.com/

Despite me being a strong advocate of open source projects, I’m convinced that the most valuable skills will live inside firms, because the valuable methods are specific. They’re the true secret sauce.

Your pass criteria, your reference script, the exact shape of your IC memo, your portfolio triage logic, the voice of your LP letter. None of that is downloadable.

That is precisely why it is worth encoding. A generic agent arrives knowing a lot about venture in the abstract. It becomes useful inside your firm only when it learns the decisions and lessons your partnership has paid years to accumulate.

I see two different paths.

Firstly, a more top-down structured path where you map your firm’s key workflows, then you shadow different professionals to understand variants thereof, to then codify and translate them into best-practice skills.

Secondly, a more bottom-up democratized path where everyone can create their own skills for whatever workflow comes to mind. If individuals feel that specific skills might be useful for others in their firm, they can submit it to a central reviewer (to ensure no personal information or structural vulnerabilities get introduced) who then makes it available for the broader firm.

While the first path has higher signal to noise ratio, the second drives more creativity and inspiration, and based on my own experience a shorter time to efficiency across the firm.

In either case, the skills can be rolled out via shared Git repos or enterprise managed settings and org-wide rollout plans.

The tempting move is to treat AI as a layer of generic intelligence sprinkled across the firm.

The better move is more boring and far more durable. Teach the machine how your firm actually works.

Your advantage will not come from the model you license, because your competitor is licensing the same one. It will come from the judgment you make reusable before they do.

Every firm already has a method. Most of it is invisible, sitting in old memos, Slack threads, reference calls, and the heads of the people who know how the work really gets done.

Skills make that method visible. A skill library turns it into the asset it should have been all along.

Stay driven,
Andre