'There just aren't enough good AI people' is the most common explanation for a stalled hiring search, and it's almost always wrong, or at least wrong as the full explanation. Good AI talent exists in numbers most searches never get close to seeing. What actually stops companies from finding it isn't scarcity. It's one of three specific, fixable bottlenecks: they're not reaching the people who exist, they're too slow to close the ones they reach, or their screening is filtering out real talent based on bad signal. Naming the actual bottleneck is the difference between a search that stays stuck and one that gets fixed in weeks.
The shortage narrative is comfortable, and mostly wrong
It's an appealing explanation because it removes fault from the process: if the talent simply doesn't exist, nobody has to examine the job posting, the screening rubric, or the six-week loop. But the volume of capable AI practitioners, people who've shipped real systems, not just completed a course, is larger than most hiring searches ever actually sample. The honest question isn't 'does this talent exist.' It's 'why isn't our search finding it,' and that question has three real answers, not one vague one.
Bottleneck one: reach
Most sourcing effort concentrates on active job-seekers: people posting 'open to work,' applying to listings, updating a profile. That's a real pool, but it's a small and skewed slice of everyone who could do the job well. The strongest AI practitioners are frequently heads-down on something that's going fine, not browsing job boards, and they will never appear in a search that only looks at people who are already looking. Reaching them requires warm signal, referrals, networks, and a partner or process that already has a relationship, not another repost of the same listing to a wider board.
- Job boards mostly surface people who are actively searching, a small and self-selecting slice of the real talent pool.
- The strongest candidates are often currently employed and not browsing listings at all.
- Referral and network-based sourcing reaches people a keyword search never will.
- A wider posting budget doesn't fix a reach problem if it's still only reaching the same searching population.
Bottleneck two: speed
Reach can be solved and the search still fails, because finding a great candidate and closing one are different problems. A six-week loop loses candidates to whichever process gets to an offer first, regardless of how good the initial match was. This is the bottleneck companies most often mistake for a shortage: they did find the person, they just weren't the ones who signed them. If your team can point to specific candidates you identified but lost to a faster offer, that's not a supply problem showing up late, it's a speed problem wearing a supply problem's mask.
Bottleneck three: signal
The third bottleneck is the least visible, because it doesn't look like a failure from the inside, it looks like 'we screened a lot of people and none of them were strong enough.' Often the screen itself is the problem: a resume filter tuned to keywords and brand-name pedigree, a coding test that measures speed under artificial pressure rather than judgment, an interview loop that rewards confident talkers over careful builders. These methods produce false negatives on exactly the practitioners who are strongest in real, ambiguous work but don't perform well on a stylized test. Bad signal quietly filters out real talent and reports back that the pool was thin.
| Bottleneck | How it looks from the inside | The fix |
|---|---|---|
| Reach | "We posted everywhere and got a weak pool" | Source through networks and referrals, not just active job-seeker channels |
| Speed | "We found good people but lost them" | Compress the loop; treat time-to-offer as a hiring metric, not a side effect |
| Signal | "Nobody in our pipeline was strong enough" | Replace proxy tests with structured evaluation of real, relevant work |
How to tell which one you actually have
Look at where candidates disappear. If strong candidates rarely enter your pipeline at all, that's reach. If they enter, impress your team, then take another offer, that's speed. If plenty of candidates enter and get screened out, but the ones you do hire underperform relative to how they scored, that's signal, your evaluation isn't measuring what it thinks it's measuring. Most stalled searches are one of these, clearly, once someone actually looks. Very few are a genuine absence of qualified people.
