Cap Table Decoded
Cap Table DecodedTypical Equity Ranges for Engineers at Seed-Stage AI Startups
Equity NormsLong read

Typical Equity Ranges for Engineers at Seed-Stage AI Startups

Hire sequence and role matter more than title when judging seed AI equity offers.

· 8 min read

Equity offers for engineers at seed-stage AI startups in 2026 need to be judged by hire sequence, role and grant terms. Older benchmarks miss both the premium for proven AI talent and the effect of higher seed valuations. A founding engineer, a later individual contributor and an ML researcher are different hires, even when their offers all say “engineer.”

Why older equity benchmarks no longer reflect what seed-stage AI startups are actually offering

An engineer comparing a live offer against a 2023 guide is using numbers from a different hiring market. AI-native startups compete with frontier labs and well-funded Series A+ companies for engineers who have shipped AI features in production. This race to hire has raised cash bands and equity floors. Carta’s H2 2025 data shows that median initial equity grants for individual contributors increased materially between 2024 and 2026, with AI/ML grants growing substantially faster than general software engineering grants.

Valuations have changed the comparison too. The median pre-money valuation for AI startups at seed is $17.9 million in 2026, which changes the valuation behind even a small ownership percentage. That makes percentage-only comparisons with older offers less useful. An outdated anchor can make a fair offer look generous or hide room to negotiate. A useful benchmark therefore needs to match the role and the point at which the engineer joins.

How hire sequence compresses equity faster than most engineers expect

Hire sequence is the first filter for any comparison. In Carta’s data, the median grant for hire #1, the founding engineer, is roughly 1.5%. Equity then compresses sharply across the first five hires. The same senior title can therefore sit behind very different grants, depending on whether the engineer joins before the product exists or after a team has begun building it. Seniority alone cannot explain that difference.

The gap reflects a change in risk. Hire #1 may join before there is a product, while hire #5 may join with Series A in sight. Both can do demanding technical work, but they are not accepting the same uncertainty. The later hire joins after some of the earliest bets have been made. A first-hire median is consequently a poor direct target for someone joining an established seed team.

In Carta’s H2 2025 report, the median seed team has roughly four employees, so each grant takes more thought about how to share the option pool. Index Ventures’ OptionPlan dataset draws on tens of thousands of grants from more than a thousand startups, providing broad context for early-hire benchmarks. But even with a large dataset, how early someone joins still makes a difference. The useful comparison remains a similar role at a similar point in the company’s life.

What a founding engineer offer actually looks like at seed in 2026

A true founding engineer is usually the first or second technical hire. The role can mean being in charge of a function, hiring engineers within a year and staying as the company grows over the long term. The scope is close to a co-founder’s. The ownership is not. Founding engineers can get as little as 0.5%, while founders usually keep a large ownership stake.

That lower-end figure and the roughly 1.5% first-hire median are reference points within the market range. They reflect differences in what the job involves, when the hire joins and how much demand there is for the candidate. Cash also deserves attention, because senior-end founding engineer base pay has moved toward near-FAANG levels as AI-native startups compete for a narrow shortlist of senior engineers. Equity remains the main way packages differ. “Founding” does not, by itself, explain whether the offer pays appropriately for the work.

Recruits Lab’s 2026 founding engineer hiring report describes a market tighter than at any point since 2021, with average time-to-fill at 6–10 weeks and rising multiple-offer pressure. That pressure gives hard-to-find candidates more room to negotiate an offer. For an engineer in the top eight percent of demand, such as a frontier-lab-adjacent candidate or a proven LLM-feature shipper, a low offer is not necessarily the final offer. The negotiating implication is direct. That leverage needs to be used before signing.

Equity for mid-level and senior ICs at seed: the range a majority of engineers actually face

Engineers who join as hires three through ten need a different benchmark from the founding hire. They are joining a funded team, so standard individual-contributor ranges apply and the upper end is lower than founding engineer guides suggest. A grant below the first-hire median can still be a good offer. It needs to fit the hire sequence, seniority and scope of the actual job. Comparing only titles skips the variable that most strongly shapes early equity.

Cash helps put that grant in context. Pre-seed and seed packages tend to trade lower base pay for more equity than later-stage equivalents. Within the US, SF/NYC sit at the top of the cash range, the rest of the country sits nearer the middle, and lower-cost metros compress base further. Location therefore affects the package being compared. A cash difference does not, on its own, establish whether the equity offer is competitive.

For an engineer with an offer in hand, the practical check is to match the offer to the right category before arguing about its size. A later senior IC offer needs to be compared with similar IC offers rather than a first-engineer median or a research scientist package. The available figures do not establish a precise numerical band for every mid-level or senior hire. Pretending otherwise would create false confidence. Published benchmarks provide a floor-check, while a credible competing offer can support a stronger ask than the benchmark alone.

Why AI/ML specialists command a separate equity band and when the premium is actually justified

“AI engineer” covers jobs with different reasons for existing. An applied AI engineer ships LLM-based product features and is the default profile for most seed companies. The compensation anchor is a base toward the middle or upper end of seed engineering pay, with equity in a range typical for seed founding hires. An ML researcher or PhD trains new models. That narrower role has a seed base anchor of $220K–$300K+, often with higher equity, when the product’s advantage depends on a model built in-house.

A company using third-party models is likely to begin with two strong product engineers, one with applied AI depth. As usage grows, an ML engineer can add evaluation infrastructure and fine-tuning. A research scientist is needed earlier when original research is at the heart of what gives the product its edge.

Leadership scope can move the comparison again. A Head of AI or Head of ML who is building the technical moat belongs closer to CTO or VP Engineering equity benchmarks than to a standard IC band. Frontier-lab alumni can also command a premium because of their established research credibility. Meanwhile, the boundary between researcher and engineer is becoming less tidy as researchers write scalable code and engineers contribute to algorithmic decisions. xAI abolished the researcher title in July 2025 and designated all technical staff as engineers.

At Series A, there are separate benchmarks for Research Engineer, Senior Research Engineer, Research Scientist and Staff Research Scientist. Pay for Senior Research Engineer and Research Scientist is above pay for Research Engineer, and Staff Research Scientist has the highest band. Those benchmarks apply to later-stage research roles. A strong applied engineer who uses a researcher benchmark can expect the wrong package, while a researcher who takes applied-engineer terms can be paid too little.

How to read a seed-stage equity offer: vesting terms, dilution, and the numbers that matter more than the headline percentage

Vesting, dilution, strike price and refresh language can tell more about its value than the headline grant size. Begin with when the equity becomes yours. The standard structure is four years with a one-year cliff and monthly vesting thereafter. The NVCA Model Legal Documents provide that default in the Investors’ Rights Agreement updated in October 2025.

A different schedule needs a close look before signing. Six-year vests, two-year cliffs and back-loaded schedules are red flags because they complicate later grants and can signal inexperience to future investors. Change-of-control terms belong in the same review. Double-trigger acceleration is standard for the first few hires and worth requesting when absent. Co-founders, advisors and some senior executives are more likely to negotiate single-trigger acceleration.

Next, model how financing dilution will change today’s percentage over time. Each funding round dilutes existing shareholders, so a sizeable seed stake may be much smaller by exit. The useful comparison is expected dollar value after realistic dilution. Strike price belongs in that calculation too. A higher company valuation does not remove the need to understand the cost and terms of the specific grant.

Restricted stock creates a separate tax deadline. Recipients have a short window from grant to file an 83(b) election, and missing it can create recurring ordinary-income tax liability as the stock vests. That issue is distinct from negotiating the initial percentage. It can nevertheless be one of the most consequential mistakes for an early hire. The grant’s legal form therefore belongs in the evaluation.

Milestone-based refreshes are starting to appear in 2026 AI engineer offers, with more grants moving from once a year to when specific technical milestones are met. Recruits Lab’s hiring report links re-grant language in the initial offer to materially reduced counter-offer impact. For the candidate, the immediate task is to find out whether that language exists before signing. A documented path to another grant is different from an informal promise to revisit equity later.

How to use these benchmarks to negotiate without leaving the offer on the table

The lesson for negotiation is to use benchmarks as a floor-check, then make the case with evidence specific to the offer. A real competing offer is the strongest lever. It shows what another company is prepared to pay, rather than only what a published benchmark says is defensible.

The negotiation should extend beyond the percentage. Ask for refresh and acceleration language in writing before signing, since generic option grant wording does not settle those terms. Ask about the option pool’s size and whether a top-up is planned at Series A. A company that has already used a large portion of an oversized pool presents a different risk from one with room left. The current 409A valuation and the latest preferred share price are inputs for assessing the grant’s realistic dollar value.

Timing limits how long that process can run. Recruits Lab’s report puts the median founding engineer candidate at 3.2 active offers at decision time in its 2024–2026 dataset and identifies speed-to-offer as a major hiring constraint. Taking too long to gather information can leave no time to decide. Candidates who enter through referrals, with the company already convinced of their fit, also negotiate from a different position because restarting the search costs the founder time. The practical aim is to resolve the grant’s terms and make the supported ask before signing, while the company is still deciding what it will take to close the hire.

Filed underEquity Norms

More in Equity Norms