How Startup Equity Actually Works for Software Engineers
Equity value depends on strike price, taxes, and dilution—not just the percentage offered.
For ML engineers evaluating opportunities at Series A and Series B AI startups, equity is a bet. This guide explains how startup equity works and why an offer only becomes readable when its grant size is weighed against exercise costs, taxes, dilution, and the chance to sell. A larger percentage does not necessarily make a better offer.
Why startup equity is harder to evaluate than engineers expect
Base salary is easy to compare because it arrives as cash. Equity can make up a significant share of compensation, yet the offer letter rarely explains enough to price it. Engineers can therefore compare the clearest part of two packages while misreading the part with the widest range of outcomes. The headline percentage is especially seductive. It seems to show how much of the company the engineer owns, though what that ownership means depends on the stage, instrument type, and tax treatment.
The central mistake is comparing percentages without knowing the strike price, the 409A valuation, or the likely dilution path. Together, they affect the value of exercising options and the cash the engineer needs to exercise them. An offer can promise large gains while leaving the employee to decide how much cash to put in, a choice that comparing salaries never brings to light.
The two instruments: what options and RSUs actually are and what each one signals
Stock options give an engineer the right to choose to buy shares at a locked strike price. The potential gain comes from the gap between that purchase price and the eventual exit price. Options require a decision before they become shares. They fit early-stage, risk-tolerant engineers who can wait for a possible payout and afford the purchase when necessary.
The two main option types are incentive stock options, or ISOs, and non-qualified stock options, or NSOs. ISOs can receive favorable tax treatment, but have strict eligibility and holding requirements. NSOs are more flexible, while their exercise spread is taxed as ordinary income. The distinction changes both the paperwork and the cash needed to participate in the upside.
Restricted stock units, or RSUs, have no strike price. Employees get the shares without buying them and pay ordinary-income tax at vesting. The quick comparison, “options are a bet, RSUs are guaranteed chips,” describes the absence of a purchase price while leaving uncertain when private-company shares can become cash. Later-stage RSUs offer more stability, while early options offer the possibility of outsized gains if the company scales.
The instrument also signals the company’s stage. Options are the most common choice from pre-seed through Series B, while RSUs are the norm at Series C and beyond. Later-stage startups and big tech companies are more likely to give annual RSU refreshes than a single grant of the kind a founding engineer gets. An ML engineer comparing Series A with Series B may thus get options in both offers. The funding label alone will not explain their different risks.
How vesting schedules, cliffs, and the 90-day window actually work
A standard grant vests over four years with a one-year cliff. Nothing vests before twelve months, when a quarter of the grant vests at once. The remaining three-quarters vest monthly or quarterly over years two through four. The offer describes a grant earned over time. It does not mean the entire amount belongs to the engineer on arrival.
For option holders, vesting is not the final hurdle. Standard grants give departing employees a fixed period, commonly 90 days, to exercise vested options or forfeit them. Someone who has earned options through years of work may still lose them because buying the shares requires more cash than they have.
At a late-stage company with a high 409A valuation, exercising can require tens or hundreds of thousands of dollars once the purchase and tax exposure enter the decision. An extended exercise window changes how much time the employee has to act. It does not turn private shares into cash.
Some companies also allow early exercise, so employees can buy shares before the options vest. When the 409A is low, the spread can be near zero, which affects the tax calculation. Acceleration addresses a different timing problem. Single-trigger acceleration speeds up vesting when there is a change of control alone, while double-trigger acceleration also needs a qualifying termination. Double-trigger terms are increasingly standard in well-structured 2026 packages.
The 409A valuation, AMT, and the tax events most engineers don't see coming
A 409A valuation appraises the fair market value of common stock and establishes the basis for option strike prices. It is usually an independent appraisal, though an in-house valuation does not receive the same safe-harbor protection. This is not the preferred price investors pay. At an early-stage startup, common stock is typically valued at a fraction of that price, so using the funding-round price as the employee’s share value blurs an important distinction.
The valuation also has a compliance purpose. Failure to meet Section 409A requirements can expose recipients to a twenty-percent federal penalty tax at vesting. For an individual offer, the immediate questions are what the current common-stock value is and how it compares with the grant’s fixed strike price. A later appraisal can rise while that strike price stays locked.
For ISO holders, that widening gap can create an unexpected tax bill. When an employee exercises and holds the shares past the calendar year, the spread between the strike price and the current 409A value becomes a preference item for Alternative Minimum Tax, or AMT. The engineer has received no sale proceeds. Even so, the tax calculation can count this phantom income while the shares still cannot be sold on an open market.
Early exercise can reduce that exposure when the spread is near zero. A zero spread leaves no AMT preference amount from the exercise. A near-zero spread should not be confused with a promise of zero tax. The associated 83(b) election is another term to discuss with a tax adviser. Each person’s taxes differ, so these rules help them work out what to ask but leave the choice of whether to exercise open.
There is also a longer holding-period question. For stock eligible for a holding-period tax benefit, the five-year holding clock starts at exercise for options and at grant for restricted stock with an 83(b) election. Selling on a secondary market before five years can forfeit that benefit. The chance to sell early therefore has to be weighed against the tax treatment attached to holding longer.
How equity percentages compress by funding stage and hire number
Grant percentages shrink as funding advances and more employees join. That reflects a change in the risk being accepted. An engineer arriving before a working product takes a different bet from one joining a company further through its funding path. The percentage becomes useful only when placed against both stage and hire number.
Pave data provides calibration points for senior founding-engineer equity in US Tier 1 markets at roughly a third of one percent at the median, roughly six-tenths of a percent at the seventy-fifth percentile, and roughly one and a quarter percent at the ninetieth percentile in the 2026 benchmarks. A candidate from a top AI lab with deep ML expertise may command the top of the range, while one without startup experience may land around the middle or below.
For a machine learning engineer comparing Series A and Series B jobs, the practical distinction is the exchange between ownership and risk. A later-stage offer generally gives up less of the company, but the smaller percentage is not automatically worse. The useful comparison is total compensation over a realistic time horizon, with illiquid equity kept distinct from salary. At Staff and Principal level in AI-native startups, compensation including annualized equity can reach the high hundreds of thousands. A large share may still be unavailable to spend.
To evaluate that exchange, an engineer should ask for the post-money valuation, dilution expectations, and realistic upside scenarios. Those figures put the grant in context. They also prevent an annualized equity number from being mistaken for the equivalent amount of dependable income.
How dilution erodes ownership across funding rounds
New funding changes the ownership denominator by adding shares. An employee can have the same grant but own a smaller percentage of the company. Early rounds such as seed and Series A usually bring fifteen to twenty-five percent dilution, which falls to roughly eight to eighteen percent in later rounds. Across three to four rounds from seed to late stage, original ownership can fall by more than half.
Consider an engineer with a tenth of a percent at Series A. If Series B and Series C each dilute existing holders by twenty percent, that stake becomes roughly six-hundredths of a percent. The employee need not leave, sell, or perform poorly. The reduction follows from how the company raises money.
Funding is not the only source of dilution. Expanding the option pool sets aside more equity for future hires and cuts the percentages that existing shareholders own. That expansion often happens before a round closes, so ownership can shrink before the new money arrives. A projection that counts only the investor’s new stake can miss this step.
Dilution does not by itself make an offer unattractive. A smaller share of a more valuable company can still produce a better outcome, but the original percentage cannot stand in for the exit percentage. Ask how many future rounds are expected, what dilution is anticipated in each, and whether the current raise includes an option-pool expansion. These are assumptions to model.
How the founding engineer trade-off differs from a standard early hire
A founding engineer gets a larger grant for taking on more kinds of work and more uncertainty. There may be no codebase, playbook, or platform team. The role can span front end, back end, databases, deployment, and sometimes design, with founder collaboration and user contact handled directly by the engineer. Technical specialization is only part of the job. In 2026, with AI writing more code, founders also want product sense and customer judgment.
For an ML engineer, this is a different career decision from taking a defined role on an established early team. Deep ML expertise can strengthen the equity offer, but it does not remove the need for technical breadth or a viable market. A large stake in a failed company is worth nothing.
Timing introduces a second trade-off. Joining after the product already has fit can remove much of the founding-engineer upside as grants compress, while preserving much of the early-stage risk. The founding label is therefore not enough to judge the exchange. What counts is the company’s actual position and the work the engineer must take on.
To judge that offer, the engineer should evaluate the post-money valuation and model two to three realistic exit outcomes after dilution and taxes. The comparison should also include unvested RSUs at a prior employer and when those shares would vest. A startup’s upside pitch competes with the candidate’s current base salary and the other compensation the candidate would leave behind.
The questions that separate a readable offer from an unreadable one
Before signing, an engineer should get answers to five questions. The first is what the current 409A valuation is and when it was last updated. The second is what the post-money valuation is and how the grant translates into dollar value today. Either figure on its own leaves the sale price uncertain.
Third, ask how many funding rounds are planned and what dilution should be modeled per round. The fourth is whether the departure exercise window is the short standard window or an extended one. The fifth is whether the company allows early exercise and what 83(b) election window applies.
A credible offer makes that math available upfront. A refusal to explain valuation, dilution, or exercise terms is a warning about how the company may treat an employee as a shareholder. Uncertainty about future outcomes is unavoidable. Opacity about the grant’s current terms is not.
The company’s equity systems also show how far it has grown up. Platforms built for this work can make accurate cap-table exports, model dilution, and let employees see the details of their grants through a portal. An engineer who can see the vesting schedule, strike price, and current 409A has a more legible promise than one who must request a founder’s spreadsheet. Software cannot make the bet pay off, but clear records let the engineer understand the bet before accepting it.
