Claude API ↗
Default model for prose, structured extraction, code, agent workflows. Sonnet for most things, Haiku when latency or volume matters.
Funds want to know what they’d be standing on. So here it is. Every tool I use in production for client work, with one sentence on why. Plus a short list of things I deliberately don’t use, because those choices matter as much as what you adopt.
Updated when my stack changes. Last reviewed May 2026.
The layer that holds the logic. Whatever runs the LLM + decides what happens next.
Default model for prose, structured extraction, code, agent workflows. Sonnet for most things, Haiku when latency or volume matters.
The right shape for letting an agent talk to your tools. I build a custom MCP server per fund — one place to govern what Claude can read and write.
Skill packaging for Claude agents. Modular, version-controlled, agent-portable. I publish to ClawTrove what I'd otherwise rewrite from scratch.
Cron, webhooks, and SaaS-to-SaaS glue when an LLM isn't the right tool. I've shipped hundreds of Zaps and still reach for it first when the job is move-this-thing-from-A-to-B. Ubiquitous integrations, dead-simple to hand off to the fund team.
Where the fund's real data lives. Everything I build either reads from these or writes to them.
The default VC CRM. I integrate at the API layer — never copy-pasting, never duplicating fields. Most of my work touches Affinity at some point.
Newer VC CRM I support alongside Affinity. Same integration shape — read deals, write enriched fields, route notifications.
When a fund needs structured data outside the CRM — deal scorecards, LP ledgers, portfolio KPI tables — Airtable is faster to build than a custom database and survives team turnover.
Source of truth for written knowledge. Memos, case notes, this site's content. The API got serious in 2024 — now it's a real backend.
When a fund needs an actual application backend — auth, custom logic, scheduled jobs — Xano is the no-code answer that doesn't lock you out.
When the fund's team has any engineering capacity, Supabase beats Xano for cost, schema control, and raw Postgres power.
The models I actually use day to day, and the wrappers around them I trust in production.
Best web research for founder/company diligence. Computer-use mode for the things only a browser session can do.
Second model. I use it for embedding, vision, and the occasional Claude failover.
Anthropic-hosted scheduled runtime for agents I don't want pinned to my laptop. Planned home for the Lab swarm once I ship it.
Where the work shows up. Static-first, edge-cached, fast.
What this site is built on. ISR + edge + good DX. Boring choice for marketing sites, which is the point.
Where I deploy almost everything. Pro plan for per-minute crons and image optimization.
When the fund needs an internal dashboard (LP reporting, portfolio KPI view, deal screener), Retool ships in days instead of weeks.
When a fund wants a marketing site they can edit themselves without engineering — that's Webflow. I don't build VC marketing sites in Webflow anymore (this site is Next), but I still ship them for portfolio companies.
The boring layer. Queues, scheduling, file storage. The stuff funds don't see but breaks first when it's wrong.
DNS, R2 storage, Workers when I need an edge function not tied to Vercel. Registrar of choice.
When something needs to run as a long-lived service (Slack bots, queue workers), Railway is the lowest-friction path.
Transactional email I trust. Pre-meeting briefs, ship notes, internal alerts.
Where most of a fund's day lives. Anything I build has to land here or it won't get used.
Where every fund I work with actually does the day. Most of my interfaces are Slack-native — Block Kit, slash commands, slack-bots over the CRM.
Microsoft Graph API for funds on the MS stack. Calendar, mail, files — everything an agent needs to brief a partner before a call.
How I onboard clients and document what I ship. Every system I deliver comes with a Loom of me using it.
Cross-post to LinkedIn + X from one place. The agent watching my social output reads from Typefully's API.
An opinionated list. If you're paying for these, we should talk before we add anything.
Self-hosted workflow tool a lot of people try to use as a Zapier replacement. In practice, you spend more time babysitting the runtime than building the workflow. I'd rather pay Zapier for glue and put real logic in the orchestrator.
Four-pillar discovery decks, 'AI transformation' workshops, three-week kickoffs. If your last call with an AI consultant started with a maturity-curve diagram, you know the one.
Most are a wrapper around Claude with a deal-scoring prompt. If you can describe what the tool does in one sentence, you can usually build it on your own stack in two weeks.
$10k / month. I deploy on the stack above, integrated into whatever your fund already uses. Book a 30-min intro and I’ll map your current setup against this list and name the three highest-leverage systems to build first.
