Guide

Building your AI product after you raise: a founder's guide

This guide is for a founder in the first year after a seed or Series A round who is building a product that uses AI. Carta's report for the first quarter of 2026 puts the median time from seed to Series A at two years, so the money has to pay for a product that earns the next round. The guide covers how long the money lasts, what it pays for, what the next investors look at, whether to use a model another company runs or train your own, what each user costs to serve, provider data terms, code ownership, security reviews and the AI claims regulators have acted on.

Published October 26, 2026. Editorial.

Key takeaways

Carta's Q1 2026 report gives a median of two years from seed to Series A and 2.1 years from Series A to Series B, so plan the spending to last past those medians.
Fine-tuning, which means training a provider's model further on your own examples, is closed to new users at OpenAI, and the Gemini API has no model that supports it. Reveneau's view is to start with a hosted model, clear instructions and evals.
OpenAI, Anthropic and Google each charge 50% of the standard price for batch requests: requests sent as a group, for work that does not need an answer at once.
The US Copyright Office says purely AI-generated material is not protected by copyright, and US law requires a transfer of copyright to be in writing and signed.
In March 2024 the SEC settled charges against two investment advisers for false statements about their use of AI, with $400,000 in total civil penalties.

In the first quarter of 2026, more than 60% of all the venture money raised by companies on Carta went to AI companies, according to Carta's report on private markets for that quarter [1]. Carta's reports count the rounds that US companies record on Carta's platform. If you have just closed a seed or Series A round for a product that uses AI, your company is in that group, and the money now has to pay for a product that earns the next round.

This guide is for that founder. It covers the first year after the money arrives: how long the money has to last, what it pays for, what investors look at before the next round, how to build the AI part, what each user costs to serve, the data terms of model providers, who owns the code, what a large customer will ask about security, and what you can safely say about your AI. Each section is short and links to a page with the detail. Every figure carries its source and its date, because prices, terms and market figures change often.

The money has to last two years or more

Carta's report for the first quarter of 2026 says the typical time between a seed round and a Series A "stands at an even two years" [1]. In a post from February 2026, Carta's Peter Walker writes that the standard advice is that venture-backed companies "should expect to raise a new round of funding every 18-24 months" [2]. A median is the middle value: half of the companies on Carta that raised a Series A took longer than two years, and half took less.

Runway is the number of months the money lasts at the current rate of spending. Burn rate is the net amount of cash that leaves the company each month, which is spending minus the cash that comes in. No source opened for this guide gives a formula, so here is this guide's own arithmetic: runway in months is cash in the bank divided by the monthly burn rate. A company with $3,000,000 in the bank and a burn rate of $125,000 has 24 months of runway. Each new hire raises the burn rate, so the same $3,000,000 lasts 20 months at a burn rate of $150,000.

So the plan has two numbers: the months of runway, and the result the next investors will want to see when those months are used up. The page how long a seed or Series A round lasts gives Carta's figures for each gap between rounds, explains bridge rounds, and shows a simple test of whether the company reaches profit on the money it has.

Most of the money pays for people

The salaries of the people who build the product are public figures you can plan with. The US Bureau of Labor Statistics (BLS) reports that the median annual wage for software developers was $135,980 in May 2025 [3]. Salary is one part of what an employee costs. In the BLS survey of employer costs for June 2026, benefits were 31.5% of the total compensation cost for full-time workers in private industry [4]. That figure covers all occupations. No BLS figure for software developers alone was found for this guide.

The page engineering budget after a raise works through the full cost of one engineer, with the payroll tax rates the IRS publishes, and sets out the other budget lines an AI product adds, such as the fees for each call to a model. For the wider comparison of hiring against working with a firm, see in-house vs outsourcing cost.

What investors look at before the next round

In June 2025, Olivia Moore and Marc Andrusko of the venture firm Andreessen Horowitz (a16z) wrote that $1 million of annual recurring revenue in the first year, once the mark of the best companies, is now at the lower end of the growth they see in AI companies [5]. Annual recurring revenue (ARR) is the revenue a company earns each year from subscriptions and other repeating contracts.

David Cahn of Sequoia Capital wrote in December 2025 that many of the best AI startups earn more than $1 million in revenue per employee [6]. That sentence was part of his predictions for 2026, and he names no data set behind it. Both statements come from investors describing the companies they see, and both describe companies that earn more revenue with fewer people than a few years ago.

The page what investors expect before the next round covers the documents a16z asks for in a data room, the warning signs it lists, and the checks lawyers list for the code in a company sale. For the full preparation, see the guide on preparing for technical due diligence.

Call a model through an API before you train your own

An AI product can send its requests to a model that another company runs. An API (application programming interface) is the way one program sends requests to another, so "calling a model through an API" means paying the provider for each request. Fine-tuning means training an existing model further on your own examples, so that it behaves the way you want with shorter instructions.

That second option is narrowing at two large providers. OpenAI's pricing page says it is "winding down the fine-tuning platform" and that the platform "is no longer accessible to new users" [7]. Google's documentation says that since Gemini 1.5 Flash-001 was deprecated in May 2025, "we no longer have a model available which supports fine-tuning in the Gemini API or AI Studio", and points to its Google Cloud platform instead [8].

Reveneau's view is that a startup should build the first version on a hosted model, called through the provider's API, with clear written instructions and examples in the request, and an eval suite that measures the result. An eval suite is a set of written, repeatable tests of the product's answers. Train or host your own model when the evals show a gap that instructions cannot close, or when a customer contract requires that data stays on servers you control. The page API or your own model for a startup sets out the three options and what each provider's terms say.

Know what each user costs before you set a price

Model providers charge per token. A token is a small piece of text, often a word or part of a word, and prices are quoted per million tokens, with separate prices for the text you send (input) and the text the model writes (output). Each large provider also sells a slower batch service for work that does not need an answer at once. OpenAI says batch requests carry a "50% cost discount compared to synchronous APIs" [9]. Anthropic says "All usage is charged at 50 percent of the standard API prices" [10]. Google says batch use "is priced at 50% of the standard interactive API cost for the equivalent model" [11].

Your gross margin is the share of revenue left after the cost of serving the product, and for an AI product the model fees sit inside that cost. The page what an AI feature costs to run builds one example from the prices read on 26 October 2026 and shows how caching and batch work change it. Investors who rebuild that margin from invoices are described in rebuilding inference gross margin from invoices.

Read the provider's data terms before customer data goes in

Every model provider writes its own rules on what it keeps, for how long, and whether it trains on it. OpenAI's documentation says: "As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)" [12]. Other terms differ between providers and between free and paid use, and a large customer may ask you about each of them. The page choosing a model provider: terms and data use puts the terms of OpenAI, Anthropic and Google side by side.

Make sure the company owns its code

The company should own what it sells, and two rules decide whether it does. The US Copyright Office wrote in January 2025: "Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements" [13]. And under US law, a transfer of copyright from the person who wrote the work "is not valid unless an instrument of conveyance, or a note or memorandum of the transfer, is in writing and signed" by the owner of the rights [14]. Code written by a contractor or a development firm needs a signed written agreement. The page who owns code a partner or an AI writes explains both rules, with the court decisions behind them.

A large customer will ask about security

A large customer may send a security questionnaire before it signs. The AICPA, the professional body of US certified public accountants, writes that "customers and business partners often request a SOC 2 report from the service organization" [15]. A SOC 2 report is a report by a firm of certified public accountants (CPAs) on a company's controls over security and related areas. The page security questions from your first large customer explains SOC 2, ISO/IEC 27001 and the Cloud Security Alliance questionnaire, and which answers depend on your model provider.

Describe your AI exactly as it works

Regulators have acted on false claims about AI. On March 18, 2024, the US Securities and Exchange Commission (SEC) announced settled charges against two investment advisers for false and misleading statements about their use of AI. The SEC's release says the firms "agreed to settle the SEC's charges and pay $400,000 in total civil penalties" [16]. In September 2024 the Federal Trade Commission (FTC), the US consumer protection agency, said its enforcement actions "make clear that there is no AI exemption from the laws on the books" [17]. The page describing your AI to investors and customers sets out each case and what it means for a pitch deck and a sales page.

The guide in one table

Question	Short answer from this guide	Page
How long must the money last?	Past Carta's Q1 2026 medians: two years from seed to Series A, 2.1 years from Series A to Series B	How long a round lasts
What does one engineer cost?	BLS median wage $135,980 (May 2025), plus payroll taxes and benefits	Engineering budget
What do investors read?	Revenue, retention by group of users, burn by month, and records that agree with each other	What investors expect
API or own model?	Start with a hosted model, instructions and evals; tune when evals show a gap	API or your own model
What does a user cost to serve?	Tokens times price, with batch at 50% at all three large providers	Running cost
What do providers keep?	It differs by provider and by free or paid use	Provider terms
Who owns the code?	The company, if contractors sign a written transfer and people write, choose or change the code	Code ownership
What will customers ask?	Often a SOC 2 report, and your model provider's data terms	Security questions
What can you claim?	Only what the product does and you can show with evidence	Describing your AI

The first 90 days

Some deadlines start the day the money arrives, and one of them is personal to each founder. A founder who receives shares that vest over time can file an 83(b) election with the IRS, which changes when the shares are taxed. The IRS form says it "must be filed no later than 30 days after the date the property was transferred" [18]. Ask a tax adviser whether it applies to you. The page first 90 days after a raise is a printable checklist, week by week, with this deadline, the SEC filing after the first sale of shares, and the product tasks that belong in the same weeks.

Where Reveneau fits

Reveneau is an AI software development consultancy with offices in San Francisco and Vancouver. Reveneau says AI writes all of its code, and every change must pass an eval suite written from the specification, the written description of what the software must do. Reveneau says Jev, TypeSafe AI's decision model, grades that suite, and that the suite runs ten times faster than it did with its previous grader. Reveneau says it uses AI instead of hiring more engineers and puts that saving into clients' prices, and that it takes responsibility for the whole project through production and after release.

For a founder after a raise, the questions in this guide are the ones to put to any development partner, Reveneau included: who owns the code, which model provider the product uses and on what terms, and what the evals show. To plan the first version of your product, see AI development at Reveneau or contact Reveneau. For how different kinds of firms suit a startup, see an AI development partner for a startup.

Explore the guide

Plan the money

How long a seed or Series A round lasts, and the time between rounds

A seed or Series A round has to last until the company can raise the next one. Carta's report for the first quarter of 2026 gives a median of two years from seed to Series A, 2.1 years from Series A to Series B and 2.4 years from Series B to Series C, for US companies on its platform. Runway is the number of months the money lasts at the current rate of spending. This page gives Carta's figures with their dates, shows the runway arithmetic, explains bridge rounds, and applies Paul Graham's test of whether a company reaches profit on the money it has.

Engineering budget after a raise: what one engineer costs, and the other lines

Each engineer adds a fixed cost to every month, so the engineering budget sets how long the money lasts. The US Bureau of Labor Statistics reports a median annual wage of $135,980 for software developers in May 2025. On top of the wage, an employer pays Social Security and Medicare tax, federal unemployment tax and benefits. This page works through the full cost of one engineer with the IRS rates and the BLS employer cost survey for June 2026, compares it with Y Combinator's 2016 figure, and lists the budget lines an AI product adds, such as model fees.

What investors expect from an AI company before the next round

Investors in the next round look at revenue, at how long customers stay, at the monthly money records, and at the code. In June 2025 Andreessen Horowitz wrote that the median enterprise AI company in its sample reached more than $2 million of annual recurring revenue in its first year and raised a Series A nine months after it started charging. This page sets out what a16z asks for in a data room and the warning signs it lists, Carta's figures for the later rounds in the first quarter of 2026, and the checks lawyers run on a company's use of open-source code.

Build the product

API or your own model: how a startup should build the AI part

A startup can put AI into its product in three ways: call a model that a provider hosts, through its API; fine-tune a hosted model on its own examples; or run an open-weight model on servers it controls. As read on 26 October 2026, OpenAI is winding down its self-serve fine-tuning and has closed it to new users, and Google's Gemini API has no model that supports fine-tuning. Reveneau's view is that a startup should start with a hosted model, clear instructions and an eval suite, and change approach only when the evals show a gap. This page sets out each option with the providers' own terms.

What an AI feature costs to run, per user, with October 2026 prices

An AI feature costs money every time a user uses it, because the model provider charges for each piece of text sent and received. This page builds one invented example, a support assistant used 40 times a month per user, and prices it with the rates OpenAI, Anthropic and Google published on 26 October 2026. On the mid-priced models the example costs $0.44 per user per month, and on the smallest models $0.022. The page then shows how prompt caching and batch requests change the figure, which charges add to it, which prices are scheduled to change, and the spending limits each provider sets.

Choosing a model provider: what OpenAI, Anthropic and Google keep and use

Before customer data goes into a model, read what the provider keeps, for how long, where, and whether it trains on it. As read on 26 October 2026, OpenAI says API data is not used for training unless the customer opts in, and keeps abuse monitoring logs for up to 30 days. Anthropic's terms say it may not train on customer content, and it deletes API inputs and outputs within 30 days with stated exceptions. Google uses prompts sent to its free Gemini API services to improve its products and keeps abuse monitoring data for 55 days. This page sets out each provider's terms side by side.

Who owns the code when a development partner or an AI writes it

A startup has to own the code it sells, and investors and buyers check that it does. Under US law, copyright first belongs to the author, an employee's work within the job belongs to the employer, and a transfer of copyright from a contractor must be in writing and signed. The US Copyright Office said in January 2025 that purely AI-generated material is not protected by copyright, while creative selection, arrangement and changes to AI output by people can be. This page explains each rule with its source, the Thaler court decision, the UK rule, and what to put in writing with a development partner.

Sell without trouble

Security questions from your first large customer, and when you need SOC 2

A large customer may check a supplier's security before it signs, with a questionnaire and a request for a SOC 2 report. The AICPA, the body of US certified public accountants, writes that customers and business partners often request a SOC 2 report from a service provider. This page explains what a SOC 2 report covers, what ISO/IEC 27001 certification is, what the Cloud Security Alliance's questionnaire asks, which answers depend on your AI model provider's data terms, and Reveneau's view on when a startup should start the SOC 2 work.

Describing your AI to investors and customers: what regulators have acted on

What a founder says about the company's AI, in a pitch deck or on a sales page, is a claim regulators can act on. The SEC has charged companies and founders over statements to investors about their use of AI, including the founder of Nate, Inc., who, the SEC alleges, raised over $42 million with false statements. The FTC has acted on claims to customers, such as Workado's AI detector advertised as 98% accurate. This page sets out the cases from 2024 to 2026 in plain words, what each claim was, and how to check every AI claim against what the product does before you make it.

Checklist

The first 90 days after a raise: a week-by-week checklist for an AI startup

The 90 days after a seed or Series A round hold some fixed deadlines and many tasks that are easier to do early. The IRS gives 30 days to file an 83(b) election after shares are transferred, and the SEC expects a Form D within 15 days after the first sale of shares in an offering that relies on it. This printable checklist puts those deadlines first, then the work that protects the company and the product: board consents, signed code assignments, model provider terms, spending limits, the runway plan, the cost per user and a check of every AI claim. Each item links to its source.

Common questions

How long should the money from a seed round last?

The money from a seed round should last past the time the next round usually takes. Carta's report for the first quarter of 2026 puts the median time from seed to Series A at two years, for US companies on its platform. A median means half of the companies took longer. Carta's Peter Walker writes that the standard advice is a new round every 18 to 24 months.

What share of venture money goes to AI companies?

More than 60% of all venture money raised by companies on Carta in the first quarter of 2026 went to AI companies, according to Carta's report on private markets for that quarter. Carta counts the rounds recorded on its own platform by US companies, so the share describes those companies and the quarter named. Later quarters can differ.

What does a software developer cost a US startup?

A software developer in the United States earned a median annual wage of $135,980 in May 2025, according to the Bureau of Labor Statistics. The employer pays more than the wage. In the BLS survey for June 2026, benefits were 31.5% of total compensation cost for full-time private industry workers, a figure that covers all occupations. Payroll taxes come on top of the salary.

Should a startup fine-tune its own AI model?

Reveneau's view is that a startup should start with a hosted model, clear instructions and an eval suite, and fine-tune only when the evals show a gap instructions cannot close. The option is also narrowing. OpenAI's pricing page says its fine-tuning platform is closed to new users, and Google says no model in the Gemini API supports fine-tuning.

Is batch processing cheaper on AI model APIs?

Batch processing costs half the standard price at all three large model providers, as read on 26 October 2026. OpenAI describes a 50% discount compared to its standard requests, Anthropic charges batch usage at 50 percent of standard prices, and Google prices batch use at 50% of the standard cost. Batch suits work that can wait for its answer.

Do AI model providers train on data sent through their API?

AI model providers each set their own rules, so read each provider's terms. OpenAI's documentation says that as of March 1, 2023, data sent to its API is not used to train or improve its models unless the customer opts in. Google's terms differ between free and paid use of the Gemini API, and Anthropic's commercial terms forbid it from training on customer content.

Can a startup own copyright in code an AI wrote?

A startup can own copyright in the parts of the code that people wrote or creatively selected, arranged or changed, and the US Copyright Office says purely AI-generated material is not protected. Its January 2025 report says each case is decided on its facts, and that creative selection, arrangement and changes to AI output by people can be protected.

When does a startup need a SOC 2 report?

A startup needs a SOC 2 report when its customers ask for one, and the AICPA writes that customers and business partners often request a SOC 2 report from a service provider. Reveneau's view is that a seed-stage company should answer its first security questionnaire honestly with the controls it has, and start the SOC 2 work when a contract depends on the report.

What happens to a company that overstates its use of AI?

A company that overstates its use of AI can face action by the SEC or the FTC. In March 2024 the SEC announced settled charges against two investment advisers for false statements about their use of AI, with $400,000 in total civil penalties. In September 2024 the FTC said there is no AI exemption from the laws on the books.

What is the 83(b) deadline for a founder?

The 83(b) deadline is 30 days after the date the shares were transferred to the founder. The IRS's Form 15620, April 2025 revision, says the election must be filed no later than 30 days after the transfer. An 83(b) election changes when shares that vest over time are taxed, so a founder should ask a tax adviser whether to file one.

What does Reveneau do for a startup after a raise?

Reveneau is an AI software development consultancy with offices in San Francisco and Vancouver. Reveneau says AI writes all of its code, every change must pass an eval suite written from the specification, and Jev, TypeSafe AI's decision model, grades that suite, which Reveneau says runs ten times faster than with its previous grader. It says it puts the saving from using AI into clients' prices.