AI Engineer Summit 2025
Insights on Building AI teams
Loading slides…
Read the talk
Insights on Building AI teams
Recruiting an AI team starts with knowing what experience matters, where talent moves, when people are ready to join, and what makes the work worth choosing.
From a talk by Heath Black
What does an Irish-literature degree prepare you to do?
What do you do with a master’s degree in Irish literature? Heath Black’s answer begins with getting creative about where to apply it. Both his sons are named after Irish writers, and storytelling eventually becomes part of his approach to recruiting. First, though, his career takes him into building products.
Black entered startups in 2009 and helped ship a conversational chatbot at Chirpify, which he describes as the first of its kind. He then worked at Imzy, an attempt to build a friendlier Reddit competitor, before joining Reddit to work on experimental business lines and trust-and-safety tools. At Meta, he worked on M, the assistant inside Messenger, and became the first product manager on the AI assistant for Ray-Ban glasses. He now leads product at SignalFire. That path supplies two perspectives on hiring: experience building AI products, and a background in understanding why stories move people.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Treat recruiting as a product problem
A product team interviews customers before deciding what to build. SignalFire applied that approach to venture capital: Black says the firm interviewed 500 founders about the problems that kept them up at night. It then built AI/ML tools and portfolio-support teams around go-to-market work, recruiting, leadership development, and product launches.
The data foundation is Beacon, SignalFire’s proprietary AI/ML platform. Black reports coverage of over 650 million employees, 80 million companies, and 200 million open-source projects. These are the platform’s claimed coverage figures at the time of the talk. SignalFire uses the information to build proprietary rankings and market insights for its own operations and the companies it backs.
For recruiting, the useful questions are more specific than how much data a platform holds: which filters identify suitable people, where those people are, when they might join, and what narrative will persuade them. Search tools can apply filters; the hiring team still has to choose the right ones.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Filter for the work, then follow the people
Academic credentials have become less prominent in AI-startup engineering hires. Black presents the following comparison, part of the hiring shift discussed in SignalFire’s research on 2023 talent trends:
| Share of AI-startup engineering hires | 2015 | 2023 |
|---|---|---|
| From top schools | 27% | 15% |
| With PhDs | 16% | 7% |
Both measures declined substantially, making an academic requirement a potentially expensive way to narrow the candidate pool.
Research roles do not make the distinction disappear. Black cites roughly 40% with advanced degrees in the research-scientist discussion; the related report describes scientist/researcher hires, rather than a census of everyone holding those titles. Advanced degrees also include more than PhDs. The practical explanation is a change in the work: foundational ML research remains important, but more hiring now concerns applying models through MLOps, product engineering, software development, and an understanding of how users interact with the result.
Employer history is changing too. Black contrasts older concentrations of AI talent at Google, Uber, Meta, and Apple with nine companies SignalFire calls the AI Ivy League. Even that map went stale quickly. He describes Inflection as having been acquired after the analysis; more precisely, Microsoft announced hiring its cofounders and several team members. The recruiting implication survives that distinction: established technology companies are now competing to attract people from the newer AI organizations.
A list of fashionable employers is less useful than a directional view of movement. In the employee-flow chart, Black highlights a positive flow from DeepMind to OpenAI and a negative trend for Cohere. Looking at both origins and destinations helps reveal where relevant experience is accumulating—and where a recruiting search may find people who have already made a similar transition.
That evidence changes the hiring process in three concrete ways:
- Inspect the body of work. For experienced candidates, examine what they built. For newcomers, look at open-source contributions and projects outside class.
- Define the actual research requirement. Decide whether the role needs a PhD researcher or an excellent engineer with relevant experience.
- Remove unnecessary academic gates. Drop or soften degree requirements when they exclude people who can do the work.
The goal is an experience-qualified top of funnel, not a credential-qualified one.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Location still shapes the candidate pool
Is San Francisco’s importance to technology hiring fading? Black tests that claim against talent concentration rather than social-media sentiment. He reports that San Francisco accounts for about 29% of startup engineers, down from 33% in 2013 but rising again since 2021. New York and Seattle, meanwhile, have each doubled their engineering market share over the period discussed. In big tech, he reports that 50% of engineers remain in the San Francisco Bay Area. These are different employment populations, so their percentages should not be treated as interchangeable.
The AI-specific picture is more concentrated. The displayed chart labels the SF Bay Area at 35% of AI employees; Black gives Seattle about 22% and New York about 10%. The Bay Area therefore exceeds the other two combined. All three locations have a larger share in the AI comparison than in the broader startup-engineering comparison, making them particularly important places to search for AI talent.
Capital concentration reinforces the pattern. Black reports that San Francisco receives nearly 38% of early-stage AI-startup funding, compared with 26% of total US early-stage funding. Distributed work has not eliminated geography: watch where people and funding move together to understand where the next candidate pool may develop.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Separate readiness to leave from readiness to join
At the airport on the way to the event, minutes separated Black from catching his flight or sitting in the lobby feeling like Charlie Brown. In baseball, a fraction of a second can separate a home run from a foul ball. Recruiting has its own timing problem, with two distinct parts: reaching someone when they may leave their current employer, and finding someone whose preferences and experience fit the hiring company’s present stage.
Retention data addresses the first part. In Black’s selected-company snapshot, Anthropic has about 66% four-year retention, while Perplexity is around 43–44%. He presents these as changing observations, not fixed properties of the companies; the talk does not specify the original cohort or observation dates. A retention curve can inform what he calls a poachability score: an estimate of when a candidate might be receptive to outreach, rather than a prediction that any particular person will leave.
Broader mobility patterns add context. Black reports that nearly 27% of Gen Z left their jobs in 2023, more than twice the Gen X rate. Within four years after graduation, he reports 2.2 jobs for Gen Z versus 1.1 for Gen X. But the explanation is not settled: slower promotions could encourage moves, frequent moves could slow promotions, and layoffs also affected the period. Black’s additional interpretation is that younger workers are more willing to take risks and bet on themselves. Neither generation nor employer retention, on its own, tells you what someone experienced at a particular company.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
An employer name needs a date
Being on the New York Knicks in 1973 means something different from being on the team in 2018. Black contrasts the championship team with a much later losing team to make a recruiting point: a company name alone does not identify the environment someone worked in. Joining before product-market fit, building the first organization, and operating a mature business are different experiences.
SignalFire’s historical composition tool reconstructs snapshots of admired companies at different points in time. Instead of searching only for an employer, a founder can ask what its organization looked like during a relevant transition: who led sales as it grew from one million to ten million, or who the first three engineers were when it shipped a key competing product. The dates connect a résumé to the work the hiring company needs done next.
Those histories help assess stage preference, risk tolerance, motivation, and the possibility of an unusually consequential hire—what Black calls a 10X hire. The practical search combines company retention patterns, changes to individual profiles, and join-and-leave dates. Outreach timing asks when someone may move; stage matching asks whether that move makes sense for both sides.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Give candidates a place in the company’s story
This is where the literature degree returns. Kurt Vonnegut’s shapes of stories place a story’s beginning and end on the horizontal axis and its movement between ill fortune and good fortune on the vertical axis. Black starts with Kafka’s Metamorphosis: Gregor Samsa wakes as a bug, and things get worse. The man-in-a-hole story traces a different path: someone falls into trouble, then climbs out.
Cinderella adds more turns. She begins in hardship, gains a magical opportunity, goes to the ball, and falls in love. Midnight abruptly reverses her fortune before the story eventually reaches lasting happiness. A recruiting narrative does not need manufactured despair, but it does need an intelligible arc: what the company has overcome, what it has achieved, why it occupies its current position, and where it is going. Pay and equity alone cannot explain that trajectory.
Compensation makes the need for a broader story especially clear. Black reports a 1.6% increase in average tech salary from November 2022 to November 2024, alongside a sharp decline in equity grants. AI hiring carries an additional premium: SignalFire’s compensation analysis, drawing on Carta data from later-stage startups, reports a 5% salary premium and a 10–20% equity premium for AI engineers over other engineering roles. A salary-only pitch makes the company compete on an advantage it may not be able to afford.
Equity is not a complete substitute. Black describes exercise participation falling to roughly 33% from 55% a couple of years earlier. The corresponding Carta report defines its Q2 2024 figure more narrowly: 32.8% of vested, in-the-money options were exercised before expiration, rather than 33% of people exercising shares. Black connects the reluctance to concerns about valuations, the cash needed to exercise, and rapidly shifting markets. Those concerns weaken equity as a standalone recruiting story.
The alternative is to identify advantages the company can actually deliver:
- Closeness and collaboration: direct work with founders and a tightly connected team.
- Speed and autonomy: less friction between deciding what to build and getting it done.
- Mission and growth: meaningful work, intellectual development, and career opportunities.
- The problem itself: complex challenges in a rapidly expanding market.
These are not interchangeable slogans. The recruiting story becomes credible when the company can explain which of them characterize the job a candidate would actually take.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Read the same signals from the other side
A team deserves the same deliberate use of evidence as a product. Experience filters define the search, geographic flows locate it, timing makes outreach more relevant, and a clear narrative connects an opportunity to a candidate’s ambitions. Black’s closing proposition is that the team is the company’s most valuable product; building it should receive comparable analytical attention.
For someone looking for work, the same evidence supports a different decision. Follow the people you admire into fields as well as companies. Watch how long they stay: tenure can offer clues about how they are treated and whether they believe the space has a future. Then decide what you want the next part of your own career arc to look like. The best-known employer is only one piece of that story.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
An embedded Vonnegut lecture and excerpts explaining how stories trace changes in fortune.
Further reading
Beacon-based research on AI hiring credentials, talent geography, and generational career mobility using 2023 trends.
An explanation of Beacon’s recruiting and investment workflows and its use of LLMs for search and data enrichment.
Historical startup salary, equity-grant, hiring, and option-exercise data, including the Q2 2024 exercise rate.
SignalFire discusses AI-engineer compensation premiums using Carta data from later-stage startups.
Updates since the talk
A subsequent report on hiring, AI-lab retention, and talent geography, with methodology for its employment data.
Full transcript
Loading transcript…