Book a time
The most advertised role in our first sweep is the AI Accelerator Architect, with 48 postings. Close behind are people who decide what a company does with AI: transformation leads, technical product managers and operating model consultants.
Where the demand is
See the figures
| Role | Postings | Employers |
|---|---|---|
| AI Accelerator Architect | 48 | 32 |
| AI Transformation Lead | 46 | 40 |
| AI Technical Product Manager | 46 | 36 |
| Research Infrastructure Engineer | 46 | 33 |
| AI Content Provenance Engineer | 44 | 38 |
| Search Relevance Engineer | 44 | 33 |
| AI Cluster Network Engineer | 44 | 31 |
| GPU Kernel Engineer | 44 | 28 |
| Online Learning Scientist | 43 | 38 |
| AI Operating Model Consultant | 43 | 33 |
We swept 377 roles and found postings for 193 of them. In total that came to 6,766 postings from 1,711 employers. The ten roles above sit at the top of that list, and they are a mixed group.
At the head is the AI Accelerator Architect: 48 postings from 32 employers. This is the person who shapes how a chip, its memory and its software fit together for AI workloads. Next to it, the GPU Kernel Engineer has 44 postings but from 28 employers, the fewest employers of any role in the ten. That suggests a small number of companies hiring several of these engineers each, which is how deep performance work tends to be staffed.
Infrastructure is well represented beyond silicon. The Research Infrastructure Engineer has 46 postings from 33 employers, and the AI Cluster Network Engineer has 44 from 31. These are the people who keep large training runs fed with data and connected across thousands of machines. Their presence says that building and running capacity is still a live hiring problem.
The other end of the list is about direction and adoption. The AI Transformation Lead has 46 postings from 40 employers, the widest spread of employers in the ten. The AI Operating Model Consultant follows with 43 postings from 33 employers. Both are about getting an organisation to change how it works once the models are available. The AI Technical Product Manager has 46 postings from 36 employers, and sits between the two ends: someone who can talk to the kernel engineers and the customer in the same week.
Three roles in the ten are about trust in what AI systems return. The AI Content Provenance Engineer has 44 postings from 38 employers. That is a broad base of employers for a role that did not exist as a job title not long ago. The Search Relevance Engineer has 44 postings from 33 employers, reflecting the work of making sure a model draws on the right material. And the Online Learning Scientist has 43 postings from 38 employers, for teams whose models need to keep learning after release.
Notice how close the numbers are. Four roles sit at 44 postings and three at 46. There is no single runaway role. The top of the list is a cluster, and a founder should read it as several kinds of demand running at once rather than one dominant hire.
Postings against employers
It is worth reading both columns. A role with many postings from few employers is a concentrated hunt, where a handful of companies are competing for the same people. A role with a similar number of postings from many employers is a wider need, and usually an easier conversation with candidates about what the job involves. The transformation and provenance roles are in the second camp. The kernel and accelerator roles lean towards the first.
By area
See the figures
| Area | Postings | Roles with postings |
|---|---|---|
| Research and learning | 1573 | 47 |
| Engineering and infrastructure | 1326 | 36 |
| Governance and society | 859 | 24 |
| Safety and security | 838 | 24 |
| Data and knowledge | 729 | 20 |
| Product and experience | 699 | 20 |
| Agents and embodied systems | 635 | 19 |
| Enterprise and commercial | 107 | 3 |
| Science and health | 0 | 0 |
| Industry and public services | 0 | 0 |
Grouped by area, Research and learning has the most postings in our sweep, with 1,573 across 47 roles. It is also the area with the most roles showing postings. Work on training, adaptation and evaluation is being advertised widely, not just by the best known labs.
Engineering and infrastructure comes next with 1,326 postings across 36 roles. That fits what we saw in the top ten: chips, kernels, clusters and networks.
Then come two areas with similar volume. Governance and society has 859 postings across 24 roles, and Safety and security has 838 across 24. Together they show that oversight, assurance and risk work is being hired for as its own discipline, not folded into engineering jobs.
Further down, Data and knowledge has 729 postings across 20 roles, Product and experience has 699 across 20, and Agents and embodied systems has 635 across 19. The agents area is the one we watch most closely at Generative Talent Co, since it covers the people who make AI systems act on a task rather than just answer. It has fewer roles than the leading areas, so month by month we will see whether it widens.
Enterprise and commercial shows 107 postings across just 3 roles. That is a small slice of our library, so it tells us little on its own, and we would not read it as a lack of commercial hiring.
Two areas, Science and health and Industry and public services, show no postings and no roles in this sweep. That does not mean nobody is hiring for AI in those fields. It more likely means the work is advertised under titles our search did not catch, or sits inside roles we have filed elsewhere. We will look at whether that changes as the sweep matures.
What stands out
Three things, none of them dramatic.
- Demand is spread across the stack. Hardware, infrastructure, strategy, product and trust all feature in the top ten.
- Some roles draw on many employers and some on few. That affects how hard they are to fill.
- The largest areas are research and engineering, but governance and safety are not far behind.
The most advertised AI jobs are not only the ones that build the models; they are also the ones that decide what to do with them.
How we read this
Each month we search recent job postings for every role in our library, including its other titles, and count the distinct postings and employers we find. This is a sample, not every job, so it is best read as direction rather than as an exact count. As this is the baseline issue, the figures stand alone, and from next month we will be able to show what moved.
Demand figures come from a monthly search of recent job postings for each role and its other titles. They are a sample, not every job in the market, and are best read as direction rather than totals.