What Jobs Are AI Startups Hiring For? November 2026 Data

If your AI startup job search stops at machine learning, you are skipping more than two-thirds of the postings in this sample. On November 3, 2026, we pulled every public posting from the job boards of 15 well-known AI companies, from Anthropic and Perplexity to Harvey, Sierra and ElevenLabs. After removing exact duplicates, 2,377 postings remained.

This article shows which functions, locations and title levels appeared, so you can aim your search. You can download the classified postings as a CSV file and check any number yourself.

What jobs are AI startups hiring for?

Engineering was the largest individual function, but it was not a majority of the postings. Sales and go-to-market, or GTM, came close behind. Here is the complete function breakdown. We classified each posting by its title first and used the department name when the title was not clear.

The overall share answers, “What portion of this entire pile of postings belongs to this function?” The median company share answers a different question: “What does the middle company look like when every company gets equal weight?” We calculate each company's function share before taking the median, including companies with no postings in that function.

That distinction matters because Anthropic contributes 677 postings, or 28.5% of the analyzed sample. A large board can influence the overall mix more than a small board. Median shares do not add up to a complete company profile because each function's median is calculated separately.

The categories describe work, not degree requirements. Research engineers sit in Engineering under our rules; research scientists sit in Research. Solutions engineering and sales engineering sit in Sales & GTM because they serve customer-facing work.

Operations includes recruiting, people work, program management, and revenue operations. Non-engineering does not mean nontechnical.

Findings you can use in your job search

Look beyond engineering without pretending the technical bar disappears. We counted 1,630 postings outside Engineering, or 68.6% of the sample. Even after removing Research too, 1,590 postings remained, or 66.9%. If your background is finance, customer success, recruiting, design, or legal, an AI-company search need not start with a career change into model research.

Build your search around the problem you can solve. A customer success candidate could show how they improved onboarding or handled an implementation that stalled. A finance candidate could show a model linking usage costs to customer revenue. These are suggestions for work samples, not claims about what every company requires.

Commercial work deserves its own search track. Sales & GTM plus Marketing accounted for 785 postings, or 33.0%. Search for customer-facing terms as well as the name of your current function. A solutions role may suit a technical candidate who prefers explaining systems and working with customers over building internal infrastructure.

Search by company mix, not just company fame. Engineering's median company share was 31.8%, while Sales & GTM's was 29.2%. That supports checking both functions across the sample rather than assuming every AI company mainly advertises research work. Use the table below to decide which boards are worth a closer read, not to rank employers by quality.

New grads need a narrower first pass. Only 29 postings, or 1.2%, carried explicit internship, new-grad, or early-career title signals. Start with those signals, then inspect unleveled roles individually. The broad search process in how to find startup jobs can help you expand beyond this selected group when its entry routes do not fit you.

Where the postings were listed

The location split below is deliberately conservative. It uses the text in the location field, not an assumption that a city listing means office-only work. “Remote North America” means the location text explicitly mentions remote or remote-friendly work alongside a US, Canadian, or North American option.

We kept mixed and unknown locations visible instead of forcing every posting into a country. A listing that offers Toronto and New York should not become a Canada-only posting because that is the location you prefer. Mexico is geographically in North America, so it has its own row rather than being counted as outside the continent.

This is not a fully remote-job rate. Some location labels say remote-friendly and may require travel. The remote flag that Ashby boards return is also too broad to establish fully remote eligibility: Ashby documents its remote flag as covering both remote and hybrid workplace types. We therefore did not promote every flagged city posting into the remote category.

Cities mentioned most often

These are city mentions, not mutually exclusive job counts. A posting offering several cities counts once for each city it mentions, so the rows should not be added together.

San Francisco includes explicit city labels, “SF Office,” and the identified company headquarters label. A generic San Francisco Bay Area label does not become a San Francisco city mention. New York combines New York City, New York, and NYC labels; it does not distinguish boroughs.

For Canadian candidates, Toronto appears in 128 postings and Montreal in 62, including mixed-location listings. That is more useful than reading the Canada-only row as the entire Canadian opportunity set. Before applying, confirm the hiring entity, work location, and required office attendance. The remote startup jobs guide covers questions to ask when a remote label leaves those details unclear.

What the title-level split says about early-career access

A title can reveal a level, but silence is not permission to assume entry level. Here is the seniority keyword split.

“No level word” includes titles that do not contain any of our level keywords. It does not mean no experience required. Likewise, “manager” can describe a product manager or account manager rather than someone who manages people.

When a title contains several signals, the script uses a fixed order: early-career terms, director or head and above, manager or lead, staff or principal, then senior. It does not treat “Member of Technical Staff” as proof of staff-level seniority. That phrase can be a job-family name rather than an experienced level.

The 1.2% early-career figure is an explicit-title proxy, not the confirmed share open to early-career applicants. Our snapshot does not include job descriptions, so we cannot measure that eligibility reliably. An internship can require graduate study, while an unleveled role can accept someone earlier in their career.

Read the experience requirements before spending time on an application. Check whether the posting asks for a specific credential, customer-facing background, shipped work, or a history of owning a system. Then compare those requirements with evidence you can show. Do not exclude yourself solely because a title is unfamiliar, but do not count it as entry level solely because “senior” is absent.

Company-by-company posting counts and funding context

The table pairs each company's posting count with its latest announced, series-labeled funding round, checked in November 2026. Each company name links to its public board or official careers page. Each round links to the announcement or reference supporting that label.

“Recorded series round” is not a claim that no later financing, extension, or tender offer occurred. Perplexity's source groups several Series E transactions and cautions about confirmation of a later extension. Cohere's Series E announcement describes an initial close. We retain those qualifications instead of treating the funding column as a simple measure of cash available.

A posting count cannot tell you whether a role is a replacement, an expansion, or an ongoing talent pipeline. Nor does a funding label tell you how the team works. If you want a small team with direct founder access, inspect the actual team structure; a familiar AI company name is not a substitute for that check. The startup versus big tech comparison can help you frame the trade-off.

How recent the recorded posting dates were

The median age of the recorded posting date was 63 calendar days. We counted 724 postings with dates within 30 days, or 30.5%, and 1,529 within 90 days, or 64.3%.

These are ages of recorded timestamps, not proof that the vacancies first appeared then. Some boards report the last update rather than the first posting date. Greenhouse's job-list response exposes an updated_at field; Ashby defines publishedAt as when the job was last published.

A refreshed description can therefore look recent without being a new opening. An older date can also belong to a posting that remained publicly listed at collection. Use recency to prioritize a board check, not to claim an applicant advantage or infer hiring speed. Confirm that a role is still listed before preparing a tailored application.

Method and limits

This is a one-day snapshot that the Unicorn Hunter team collected for this article. It does not measure the product's wider coverage. We pulled the public boards on November 3, 2026: 13 run on Ashby and two on Greenhouse. Live counts will differ from the snapshot. The files are public: the classified postings (CSV), the summary results (JSON) and the analysis script (Python).

We removed duplicates only when company, title, and location matched exactly. That removed four rows. We also planned to include Runway and Hebbia. The Runway board in our list turned out to belong to a different company, CFO.ai, so we excluded its three rows. We did not find a public board for Hebbia. Neither company is in the counts. The same title in a different location remains a separate posting. Multiple requisitions with identical labels can be collapsed by this rule, so even the cleaned count is not a count of unique seats.

The script applies ordered keyword rules to titles, then uses department text if the title has no match. Finance and legal terms receive specific categories. Recruiting, operations, and program management go into Operations & Strategy.

Design and analytics have separate rules; sales, partnerships, solutions, and customer success go into Sales & GTM. Unmatched titles remain Other. The published results include the complete patterns and category decisions.

We manually read a random sample of 60 classifications. Agreement with our saved labels was 54 out of 60, or 90.0%, before corrections; it was 60 out of 60, or 100.0%, afterward. Fixes addressed titles such as Salesforce Developer and industry wording such as “Financial Services” in the snapshot.

The summary results file keeps the initial rules and the labels, so you can reproduce the comparison. The corrected score reuses the same sample; it is not an independent accuracy estimate for the whole dataset.

Posting counts are not hiring counts. The selected companies are not a random sample of AI startups, and this snapshot does not establish a trend over time. It mixes functions, countries, and company sizes, including mature private companies.

Location text does not settle remote eligibility, and title words do not settle experience requirements. Use these results to choose where to investigate, not to estimate your odds of an offer.

Next steps

• Choose a function before choosing a famous employer. Compare your evidence with the work in Engineering, commercial teams, operations, or another category. Keep technical requirements visible even when the function is non-engineering.
• Open the company boards that match your function. Read the team descriptions and actual experience requirements. Check whether an apparent cluster reflects the same role advertised in several places.
• Resolve location and level early. Confirm office attendance and hiring geography. If you are early-career, start with explicit entry signals and evaluate unleveled roles individually.
• Prioritize recent timestamps, then verify the posting. A date is a useful sorting aid, not proof of a new vacancy. Tailor your resume to the role you confirmed, not to a count in this snapshot.