In partnership with
Last Tuesday, a friend at a mid-tier venture fund pinged me on Slack. "We just deployed a multi-agent system that cross-references our data rooms with public filings and founder LinkedIn profiles. It flagged three inconsistencies in a Series B deck that the associate missed." She's not an engineer. She's a former consultant who learned to prompt-chain Claude into a due diligence pipeline over a long weekend.
I sat with that for a minute. The tech didn't surprise me. The org chart implications did.
85% of dealmakers use AI daily. The other 15% should be worried.
A survey of nearly 300 private capital dealmakers found that 85% now use AI to automate daily tasks, up from 76% a year ago. GoingVC published a breakdown in February that split the industry into two camps: AI-first firms and AI-later firms. AI-later firms bolt tools onto manual workflows, maybe shaving a few hours off a memo. AI-first firms rebuilt the whole pipeline from scratch. Sourcing runs through Harmonic and Specter, tracking engineering velocity and founder signals before a partner ever sees a name. Data rooms get parsed by Hebbia and Claude in parallel, cross-referencing hundreds of documents for internal consistency. Meetings get transcribed by Granola and Fireflies, structured into searchable records, and fed back into the CRM.
If you're an engineer, you recognize this immediately. It's the difference between writing a script to automate one step and redesigning the system architecture. One compounds. The other doesn't.
The analyst job is shapeshifting
I want to be careful here because "AI replaces analysts" is easy and wrong. What's happening is weirder and more interesting. Bessemer reclaimed 234 hours per analyst after integrating AI into their workflows. BlackRock saw a 5x increase in research throughput. Eximius Ventures built an AI-powered deal funnel that automates initial filtering and scoring, then hands off to humans for the stuff that requires taste.
The old analyst job was: read decks, build comps, write memos, update the CRM. The new one is: configure the AI pipeline, validate its outputs, catch the false positives, and spend your time on the 20% of decisions where relationships and judgment actually matter. Six weeks of diligence compresses into four hours of AI review followed by focused human analysis. 100% of data room materials get examined instead of the old 15%.
One NLP system caught problematic contract terms in 87% of cases where issues later surfaced, compared to 63% for manual review. If you came from engineering, this should feel familiar. You're not doing less work. You're the person who builds the CI/CD pipeline instead of running tests by hand.
LLM traffic converts 3× better than Google search
58% of buyers now start their research in ChatGPT or Gemini, not Google. Most startups aren't showing up there yet.
The ones that are get cited by the AI tools their buyers, investors, and future hires already use. And they convert at 3×.
Download the free AEO Playbook for Startups from HubSpot and get the exact steps to start showing up. Five minutes to read.
The SWE-to-VC pitch just changed
When I made the jump from engineering to investing, the pitch was simple: I could read code and sniff out technical debt in a diligence call. The bar was low because most investors couldn't do it.
That bar moved. Fast.
The AI-native VC firm needs someone who can build the firm's own tech stack, not just evaluate a startup's. The tools exist: Affinity for network mapping, Tactyc for portfolio analytics, Standard Metrics for reporting. But stitching them into a system that compounds knowledge over time? That's an engineering problem. Most VCs still can't solve it.
I've talked to three firms in the past month hiring "internal tooling" roles. At a VC fund. People building Notion integrations and LLM pipelines for a 15-person investment team. Two years ago these roles didn't exist. Now they're the cost of entry if you want to compete on speed.
What happens when the tools outrun the toolmakers
So what happens when you don't need the engineering background to use any of this? My friend, the former consultant, built her multi-agent diligence system without writing a line of code. Claude and a no-code workflow tool. Done. The moat I thought I had (technical fluency) is eroding the way every moat erodes: the tools get cheaper until everyone has them.
I don't have a clean answer. The optimistic read is that engineering judgment compounds in ways that prompting doesn't, that knowing why a system fails matters more than knowing how to query it. The pessimistic read is that I'm telling myself a comforting story while the ground shifts under me.
The SWE-to-VC path isn't dying. But the job description changed faster than I expected. And if you can't build the system that reads the code, someone with a weekend and a Claude subscription might get there first.
I'm updating my own firm's pipeline this weekend. Not because I want to. Because the 15% who aren't using AI yet? I don't want to find out what happens to them.
SWEdonym