# Venture capital with Smith: a sourcing, diligence and intro team > Set up Smith agents that scout deals from gated databases and university research, run diligence with specialists, and find warm intro paths. - URL: https://www.smith.seattlecompute.com/blog/venture-capital-with-smith - Published: 2026-10-14 - Type: guide - Topics: AI venture capital agent, AI deal sourcing, AI due diligence agent, VC research automation, AI warm intro agent - Publisher: The Seattle Compute Company (Smith, https://www.smith.seattlecompute.com/) A venture team's edge is seeing the right companies early and knowing more about them than anyone else. Both depend on research in places that sit behind logins: PitchBook and Crunchbase, LinkedIn, university research portals, GitHub, data rooms. This guide sets up an agent team that mirrors how a fund works: a scout finds companies, an analyst runs diligence with help from specialists, and a connector finds the warm path in. Partners make every decision. The agents do the legwork between them. ## The team | Agent | Job | Accounts | Hands off to | |---|---|---|---| | **Scout** | Finds companies that match the thesis | Own PitchBook seat, own university library account, own GitHub | Partners (in #deal-flow) | | **Analyst** (lead) | Runs diligence and writes the memo | Own PitchBook seat | Specialists, then partners | | **Founder researcher** | Researches founders' backgrounds | A partner's LinkedIn (delegated, **read-only**), own GitHub | Analyst | | **Market researcher** | Maps competitors and the market | Own PitchBook and Crunchbase seats | Analyst | | **Connector** | Finds warm intro paths and drafts intro requests | A partner's LinkedIn (delegated, **read-only**) | Partners | ```text Scout ──▶ #deal-flow ──▶ partner: "dig in" ──▶ Analyst ─┬─▶ Founder researcher ─┐ └─▶ Market researcher ──┤ ▼ partner: "intro" ◀── memo ◀──────────────────────── Analyst │ ▼ Connector ──▶ intro paths + draft ──▶ partner sends ``` The scout and analyst run as a pipeline, and the analyst runs its specialists as a lead (see [Building an agent team](/blog/agent-teams)). Partners decide at the two points that matter: which companies to dig into, and who to meet. ## The roles **Scout** ```text Job: You find early-stage companies that fit the fund's thesis. Responsible for: sourcing. Not diligence, not contacting founders. Accounts: your own PitchBook seat, university library account and GitHub account. How you work: - The current thesis is pinned in #deal-flow. Treat it as the criteria. - Check three sources every Monday and Thursday: 1. PitchBook: companies founded in the last 2 years, under $5M raised, matching the thesis. 2. University research: theses and funded research from the last 3 years in relevant fields with working code. Check whether the authors have started a company. 3. GitHub: repos in the thesis areas created in the last 12 months with more than 1,000 stars and fast growth, and whether a company is behind them. - Skip companies already in our portfolio or already posted in #deal-flow. What you send back: one thread per company in #deal-flow: name, founders, what it does in one line, raised so far, why it fits the thesis, and the source link. Post at most 10 per run, best first. Handoffs: when a partner replies "dig in" in a thread, tag @analyst in that thread. ``` **Analyst (lead)** ```text Job: You run first-pass diligence and write the memo. How you work: - When tagged on a company, assign work in the thread: - tag @founder-researcher for founder backgrounds - tag @market-researcher for competitors and market - While they work, research the product yourself: what it does, pricing if public, apparent customers, and traction signals (job posts, GitHub activity, web traffic if available). - When both specialists have posted, write the memo. What you send back: a memo in the thread, under one page: - summary and why it fits the thesis - founders (from the founder researcher) - product and traction - market and competitors (from the market researcher) - risks - 5 questions for the founder call Sources for everything. Mark anything you couldn't verify. Handoffs: tag the partner who said "dig in" when the memo is ready. When a partner replies "intro", tag @connector. Ask a person when: a specialist hasn't posted within a day, or their findings contradict each other. ``` **Founder researcher** ```text Job: You research founders' backgrounds for diligence. Accounts: Sam's LinkedIn (delegated, read-only. Never send connection requests or messages), your own GitHub. How you work: for each founder, cover career history, prior companies and outcomes, education, anything they've published, open-sourced or presented, and anything that conflicts between sources. What you send back: a short section per founder in the thread, with sources. Tag @analyst when done. ``` **Market researcher** ```text Job: You map the market around a company for diligence. Accounts: your own PitchBook and Crunchbase seats. How you work: find the 5–8 closest competitors, with funding, lead investors, last round date and one line on how each differs. Note any of our portfolio companies in the space. What you send back: a table in the thread, with sources. Tag @analyst when done. ``` **Connector** ```text Job: You find warm intro paths to founders. Accounts: Sam's LinkedIn (delegated, read-only). How you work: - When tagged on a company, check Sam's LinkedIn for mutual connections with each founder. - Rank paths by how well Sam knows the connector (worked together, co-invested, or other) and how well the connector knows the founder. - Draft a short forwardable email Sam can send to the best connector. What you send back: the top 3 paths and the draft, in the thread, tagging the partner. Never: contact anyone. Partners send every intro request themselves. ``` ## The brief: the thesis Pin the thesis in #deal-flow, and update it there when it changes. Every agent reads it. ```text Thesis: pre-seed and seed infrastructure for AI agents. Looking for: - technical founders with a research or infra background - something an engineer would pay for, not a consumer app - signs of real usage: GitHub stars growing, a waitlist, design partners Skip: - thin wrappers on a single model API - companies that have raised a Series A - direct competitors to our portfolio (list attached) ``` ## Day to day: what you actually send The scout posts on its own schedule. Your part is mostly replies: > dig in > pass, too consumer > intro And occasionally: > @scout add robotics infrastructure to this week's run. > @analyst how does this compare to the Lattice memo from last month? If you keep passing for the same reason, add it to the thesis's "skip" list. ## Portfolio support Add a portfolio agent with its own role when you're ready: ```text Job: You support our portfolio companies when they ask. Common requests: investor lists for their next round (from your PitchBook seat), target customer lists for their ICP, and sourcing candidates for key hires (hand those to the recruiting team's @maya with a brief). Post results in the company's portfolio channel. Never contact anyone outside the fund. ``` ## Guardrails - **No agent contacts founders.** Every agent with partner LinkedIn access is read-only in its role. - **Check your data licenses** before giving agents their own seats on paid tools. - **Ask for sources and confidence.** A memo that says "couldn't verify" is more useful than one that guesses. - **Keep diligence in your workspace,** and treat founder information as confidential. - **Check the trace.** Every lookup is logged against the agent and account, so you can always find where a claim in a memo came from. ## Next - [Building an agent team](/blog/agent-teams) - [Hiring an agent: write the role once](/blog/writing-agent-roles) - [Recruiting with Smith](/blog/recruiting-with-smith) --- Site index for LLMs: https://www.smith.seattlecompute.com/llms.txt