Home › Blog › Founding an AGI Startup, Part 3
Founding an AGI Startup · Part 3 of 11 👥

The Founding Team — Co-Founders, Early Hires, and Equity

Legal structure protects a team that already exists. This article is about building that team in the first place — who you actually need on day one, how equity splits really get decided, and why Anthropic's seven co-founders chose near-identical shares over a "who gets the most" negotiation.

FL
FrontierAGI Team
Startups Team Building Simulation
What this series is. Part 3 of a standalone founder-playbook simulation, run tri-lane per reader direction — every step below covers frontier scale-first, applied/agentic-layer, and narrow research-bet companies in parallel. Educational simulation content, not legal or HR advice.

1. 🧭 Why the Team Question Comes Before the Product Question

Part 2's legal structure protects and formalizes a team — it doesn't tell you who should be on it. That's a harder, more consequential decision, and the data on startup failure backs this up repeatedly: co-founder conflict and team dysfunction are among the most commonly cited reasons early companies collapse, ahead of running out of money or building the wrong product. In an AI-specific company, this gets sharper: the skill sets you need (ML research, infrastructure engineering, product, distribution) are unevenly distributed, expensive, and — for the research-heavy roles — genuinely scarce relative to demand. Getting the team question right, lane by lane, is this article's entire job.

7 Founders
Anthropic's founding team size — an unusually large number for a company this successful, and a deliberate choice, not an accident
~Equal Shares
Anthropic's founder equity split — near-identical stakes across all seven, rather than a CEO-takes-most default
1-2 People
Realistic founding team size for most Lane 2 companies at the idea-validation stage — smaller than founders often assume they need
#1 Killer
Co-founder conflict's ranking among commonly cited causes of early startup failure, ahead of running out of money

2. 🤝 Choosing Co-Founders

🔴 Lane 1: Frontier Scale-First

You need co-founders whose combined credibility can unlock nine-figure capital before any product exists — in practice, this usually means at least one founder with a genuine track record at a frontier lab (research leadership, not just an engineering role), because investors writing checks this large are underwriting the team's ability to compete for scarce research talent as much as the idea itself. A common, realistic pattern is a research-credible co-founder paired with an operations/fundraising-focused co-founder — splitting "can this team actually build a frontier model" from "can this team actually run a company at this capital scale," since very few individuals are genuinely strong at both.

🔵 Lane 2: Applied / Agentic Layer

Most Lane 2 companies are genuinely fine starting with one or two founders — a technical founder who can build the initial product and a founder (sometimes the same person) who can talk to customers and sell. The classic complementary pairing here is technical-plus-commercial, but the more important test isn't skill complementarity on paper, it's whether you've actually worked together before under real pressure. A co-founder relationship formed purely from a pitch-competition introduction carries meaningfully more risk than one built on a prior working relationship, regardless of how well the resumes match on paper.

⚪ Lane 3: Narrow Research Bet

Here the co-founder question is almost entirely about research credibility and shared technical conviction in a specific, currently-unproven thesis — investors in this lane are betting on the founding researchers' judgment about an open problem, so a founding team without deep, personally-held conviction in the specific research direction (not just general AI enthusiasm) is a weak pitch regardless of individual pedigree. A non-research co-founder handling operations and fundraising is still valuable, but the research conviction has to be genuinely shared among the technical founders, not delegated to one person while others go along.

3. 🎯 Early Hires — Who You Actually Need First

🔴 Lane 1: Research Infrastructure Engineers
Before your fifth research scientist, you need people who can build and operate distributed training infrastructure reliably — a shortage of this specific skill set, not research talent alone, is a common bottleneck for well-funded but early scale-first teams.
🔵 Lane 2: A Second Engineer, Then Sales
The realistic first few hires are almost always an additional engineer to increase shipping speed, followed by a dedicated sales or customer-success hire the moment founders can no longer personally handle every customer conversation — hiring "growth" or "marketing" roles before product-market fit is a common, avoidable early misstep.
⚪ Lane 3: A Small, Senior Research Core
Research-bet companies tend to stay deliberately small and senior-heavy for longer than the other two lanes — a handful of highly experienced researchers with real conviction in the thesis, rather than a larger team of more junior hires, since the work is exploratory enough that coordination overhead across many people can slow progress rather than speed it.

3b. 🗂️ What the Team Actually Looks Like, Role by Role

Putting Sections 2 and 3 into a single reference: here's a realistic team structure at roughly the seed stage (first 12-18 months) for each lane — not a hiring plan to execute mechanically, but a sense of scale, seniority mix, and where headcount actually concentrates.

Role🔴 Lane 1: Scale-First🔵 Lane 2: Applied Layer⚪ Lane 3: Research Bet
CEO / Fundraising LeadFounder, usually with prior frontier-lab or exec credibilityFounder, often the commercial/product co-founderFounder, usually a senior researcher who also carries the pitch
Research / Technical LeadFounder or very early hire with published research track recordOften the same person as CEO, or a technical co-founderOne or more founders — the entire pitch rests on this credibility
Engineering (headcount)Heavy — infrastructure/training engineers dominate early hiringLight — 1-3 generalist engineers total pre-Series ALight — engineering only supports the research, doesn't lead it
Research staff (headcount)Growing fast, often 10+ within a year if capital allowsUsually none — no dedicated research functionSmall, senior, deliberately slow-growing (Section 3)
Product / DesignMinimal or absent pre-launchCritical — often a founder wears this hat directlyAbsent — no product exists yet by design
Sales / Customer SuccessNot needed pre-revenueFirst non-engineering hire, arrives earlyNot applicable — no customers yet
Operations / FinanceDedicated hire fairly early — capital deployment at this scale needs real financial controlFounder-handled or outsourced (bookkeeper/fractional CFO) until Series AFounder-handled or outsourced, similar to Lane 2
Total headcount at $2-5M raisedOften still under 15 — capital goes to compute, not headcount, at this stage3-8 peopleUnder 10, senior-weighted

Illustrative structure for the simulation's assumptions (Part 1, Section 6) — actual team shape varies by specific idea and founder background.

4. 💰 Equity Splits — How They Actually Get Decided

The default advice — "split equity based on contribution" — is true and almost useless in practice, because contribution is nearly impossible to measure fairly at the moment a company is founded, before anyone knows what the real work will turn out to be. Two structurally different approaches show up repeatedly in real companies:

⚖️ Equal (or Near-Equal) Splits
Common among founding teams with roughly comparable seniority and shared conviction in the mission — it sidesteps a contentious negotiation at the exact moment trust is most important to establish, at the cost of not reflecting differences in role or risk that may emerge later. Works best with strong shared values and a genuine peer relationship among founders.
📊 Weighted by Role, Risk, and Timing
More common when founders join at different times, take on the CEO/legal-risk role, or bring disproportionate resources (capital, an existing customer base, critical IP) — a founder joining six months after incorporation with an already-built prototype typically takes a smaller share than the person who took the initial risk, and this should be explicit and agreed upfront, not negotiated after tension already exists.

Regardless of which split you choose, vesting is non-negotiable — standard four-year vesting with a one-year cliff protects the company (and remaining co-founders) if someone leaves early, and its absence is one of the most common, most regretted early-stage mistakes across every lane.

5. 🏛️ Case Study: Anthropic's Seven-Founder Equal Split

Anthropic
Founded 2021
7 co-founders All from OpenAI Near-equal equity stakes
Dario Amodei (former OpenAI VP of Research) and his sister Daniela Amodei (former OpenAI VP of Safety & Policy) founded Anthropic in 2021 alongside five additional OpenAI colleagues: Tom Brown (lead author of the GPT-3 paper), Jared Kaplan (author of the neural scaling laws research), Chris Olah (an interpretability research pioneer — directly relevant to the field covered in this site's interpretability deep dive), Jack Clark (former OpenAI policy lead), and Sam McCandlish. All seven left together, reportedly over shared concerns that OpenAI's priorities were shifting away from safety research toward faster commercial deployment — a genuinely shared conviction in a specific thesis, not a single founder's idea that others joined later. The equity structure reflects that: rather than a conventional CEO-takes-the-largest-share default, all seven founders hold near-identical stakes (each under 1% today, following the company's growth and dilution across many funding rounds) — an unusually flat structure for a company of Anthropic's eventual scale and valuation.
The lesson for this article: a large founding team is a real risk (more people who need to agree, more equity to divide, more potential for future conflict) but Anthropic's case shows it can work specifically when the founders share deep, pre-existing conviction in one thesis and choose a flat equity structure that avoids a status hierarchy among people who see themselves as peers — the equal split wasn't a failure to negotiate, it was a deliberate signal about how the company intended to make decisions going forward.

5b. 🧭 Do You Need a Board, Advisors, or Mentors This Early?

Founders often assume a formal board is either irrelevant at seed stage or mandatory from day one — neither is quite right, and the honest answer differs by lane and by what stage of capital you're at.

🔴 Lane 1: Frontier Scale-First

A real, active board arrives early and matters immediately, because the capital amounts involved mean investors will require board seats as a condition of funding, not a courtesy. Beyond formal governance, this lane benefits most from senior technical advisors — ideally people with direct frontier-lab research leadership experience — who can pressure-test research direction and compute-allocation decisions the founding team may not have faced before at this scale. A board that includes at least one member with genuine hands-on experience scaling a research organization (not just a generalist VC board seat) is a real, practical asset here, not a formality.

🔵 Lane 2: Applied / Agentic Layer

A formal board is usually unnecessary before an institutional seed or Series A round — most Lane 2 founders operate with no board at all in the earliest months, just founders making decisions directly. What's genuinely useful this early is a small, informal group of 2-4 advisors: ideally someone who has built and sold a similar product, and someone with domain expertise in your specific customer vertical. This can be structured with small advisor equity grants (commonly 0.1-0.5% each, vested over one to two years) rather than a formal board seat — lower commitment, easier to adjust if the relationship isn't adding value. A formal board typically forms once an institutional investor's term sheet requires one, discussed further in Part 4.

⚪ Lane 3: Narrow Research Bet

Similar to Lane 1 in that investor-required board seats arrive early given the capital scale, but the more valuable addition here is often a small scientific advisory group distinct from the governing board — respected researchers in the specific subfield who can vouch for the thesis publicly and sanity-check research direction privately, without holding formal governance authority. This matters more in Lane 3 than either other lane because the company's entire credibility rests on a technical bet outsiders (investors, future hires, eventually the public) have to trust before any product exists to prove it.

A board seat is a governance obligation you take on for capital. A mentor or advisor is a relationship you choose because it makes the company better — conflating the two, in either direction, is a common early mistake.

6. 📋 Side-by-Side: Team Building by Lane

Factor🔴 Lane 1: Scale-First🔵 Lane 2: Applied Layer⚪ Lane 3: Research Bet
Ideal founding team size3-7, mixing research and operational credibility1-2, technical + commercial2-5, senior research-heavy
First non-founder hiresResearch infrastructure engineersSecond engineer, then salesAdditional senior researchers
Equity split tendencyWeighted by seniority and fundraising roleWeighted, sometimes equal for close peersOften near-equal among research founders
Biggest hiring constraintCompeting with incumbent labs' compensationCash-constrained hiring before revenueNarrow pool of researchers who share the thesis
Team size at $2-5M raisedStill pre-revenue, team may already be 20+Typically 3-8 peopleOften still under 10, senior-weighted

7. ⚠️ Risk Flags

🤝
Co-Founding With Strangers Under Time Pressure
Meeting a co-founder days before incorporating, driven by urgency to "just start," is a recurring pattern in failed founding teams across all three lanes — a prior working relationship, even informal, meaningfully reduces this risk.
📜
No Vesting, No Buy-Sell Agreement
A co-founder who leaves after three months while retaining a full, unvested equity stake is one of the most common and entirely preventable early-stage disputes — this is what Section 4's vesting requirement exists to prevent.
🏗️
Over-Hiring Before Product-Market Fit
Most acute in Lane 2 — hiring a large team on the strength of a seed round before validating real customer demand burns runway on payroll rather than on the iteration needed to actually find fit.
🎭
Manufactured Research Conviction
Specific to Lane 3 — a founding team that adopts a research thesis because it's fundable, rather than because they deeply believe it, tends to fracture under the real pressure of years without a shippable product; investors and later hires can often tell the difference.
Anthropic's seven founders didn't negotiate an equal split — they already agreed on the thing that mattered before they incorporated, and the equity structure just reflected that agreement honestly.

8. 🧪 Team-Building Checklist (All Three Lanes)

1
Have you actually worked together before? Not required, but its absence is a real, quantifiable risk factor you should account for explicitly, not ignore.
2
Do all co-founders genuinely agree on the core thesis — not just the general opportunity, but the specific lane and approach — before any equity conversation happens?
3
Is the equity split explicit, documented, and vested — four years, one-year cliff, no exceptions — regardless of whether the split itself is equal or weighted?
4
Have you identified the actual next 1-2 hires your specific lane requires (Section 3), rather than hiring generically "good people" without a clear gap they fill?
5
Does your team composition match what investors in your lane will actually expect to see — research credibility for Lanes 1/3, product-and-distribution capability for Lane 2 — before you're in the room pitching them?

9. 🧭 What's Next in the Series

Part 4 covers Seed Capital — pitching, valuation, and term sheets, again across all three lanes: what investors in each lane actually want to see, how valuation gets set differently for a pre-revenue product company versus a research-thesis company, and what a realistic term sheet looks like at each capital tier.