We instrument venture capital
Scale has run an internal engineering team for years. Our investment team works every day on systems we built ourselves: a research platform with 41 tools, eight autonomous agents, and scoring models trained on eight years of funding history. This page is the inventory.
By the Numbers
57
Active repositories
41
MCP tools
08
Autonomous agents
08
ML pipelines
.webp)
The tech stack
The research platform
One MCP server and REST API behind every AI surface at the firm. It fronts our data warehouse, web agents, CRM, and more, resolves fuzzy company names to real records, and runs semantic search over embeddings. Every write tool previews before it commits and is labeled by blast radius, so a model can't commit a change no human has reviewed.
Autonomous research agents
Background agents that take a research job and return with the work done, detailed below.
Sloppy Joe
Feedback from any internal tool goes straight to an agent that reproduces the issue, writes the fix, and opens a draft PR overnight. It also sweeps for tech debt and keeps architecture docs current on its own.
Scale Signal
An internal chat interface that streams every tool call live, remembers users across sessions, and gives each one a sandboxed Python interpreter and real workspace apps.
Scale Studio
The portfolio-monitoring platform the deal team opens every day: financials, cap tables, and fund waterfalls across four microservices, syncing bidirectionally with Salesforce.
Data & ML pipelines
Monthly scoring of every tracked company by four stage-specific models trained on funding data back to 2018, plus Chronocluster, our temporal clustering library that keeps the market taxonomy stable week over week.
A skill for every workflow
All common workflows for investors packaged into reusable and shared skills that automatically improve and learn from usage.
Chrome extension
Flags whetherthe site you're on is already in the CRM, drafts outreach emails inline
in Gmail, and turns LinkedIn profiles and DocSend decks into CRM records in one click.
Business intelligence
Dashboards over pipeline, market, and portfolio data, built on the same warehouse every pipeline on this page reads from and feeds.
Deal voting
Where the partnership votes on deals, with AI-generated round summaries so every vote starts from the same picture of the company.
Data labeling
A company-taxonomy labeling queue where the investment team's judgments become training constraints for our clustering and scoring models.
LMS
An internal learning platform where the team trains on the tools above, from MCP basics to building their own agents.
Agents on the team
These tools hand back a run ID and go to work in the background. Each is a small, purpose-built pipeline
with hard rules about what it may and may not do.
01
Stealth founder research
Synthesizes who a founder is and what they might be building from public signal, so we can form a view and reach out while a company is still quiet, before the round is competitive.
02
Researcher quality
Scores a person as a technical researcher across six dimensions and finds warm paths through co-authors in our network, so deep tech diligence rests on evidence instead of pedigree.
03
Competitor research
Identifies a company's closest competitors from public signal alone, so founders get a sharper read on their market from the first meeting and our conviction is built on the full field.
04
Market research
Builds a cited bull/bear report reviewed by a critic promoted as a strict investment-committee reviewer, so every deal discussion starts from evidence both sides can check.
05
Customer intro finder
Finds warm paths from our network to the buyers a portfolio company is trying to reach, because the fastest help we can give after a check is a customer.
06
Hiring recommendation finder
Surfaces candidates for portfolio companies open roles from our network and hiring signal across the market, because the second-fastest help is a great hire.
07
Job postings research
Reads a company's live job postings directly from its ATS, because who a company is hiring is one of the most honest signals of where it's actually going.
08
Diligence content
Writes a founder-meeting packet around an explicit thesis of what would have to be true for this to be a great investment, so meetings go deep instead of covering basics.
How we built it
Explore all perspectives
Open source
We open source the parts of our stack other teams can use.
Open source is where much of the best infrastructure of the last two decades came from, and it's a pattern we back as investors: companies built on open code, open standards, and the communities around them. We contribute the same way — tools we built for our own engineering and agent workflows, released for anyone to use at
Windbag
A linter that catches comments narrating change history instead of documenting the constraint that makes the code correct today. Runs as an agent hook org-wide.
Swamp-claude-log-digest
A swamp club extension that distills reusable skills from Claude Code session logs, so what one engineer figures out once becomes something every agent knows.
Circleci-orb-version-bump
A CI guard that fails a build when a version wasn't bumped against the base branch, built for the era of AI-agent pull requests.
Riff
A Jev-based linter for writing quality.
Agents at Scale Fellowship
Every semester, we bring on one undergraduate or graduate student of any major, who can point to applied AI work they've already built — not just coursework. Comfortable with ambiguity, motivated by a real deadline, and able to work part-time and remotely alongside a full course load. Our first Fellow built a competitive-intelligence agent the investment team uses today.
Fall 2026 cohort
Applications for the fall semester are open now. We're moving quickly: apply by September 25, 2026, for a fixed-term engagement running October through mid-December
