Hey everyone, Neil here. You're reading High-Signal Hiring. Hiring systems from 20+ years of global recruitment experience and 500+ technical hires. Zero noise and instantly actionable.
Last issue we covered AI interview fraud. Cluely overlays, deepfake personas, proxy candidates, and how to redesign the loop so AI assistance becomes self-exposing.
This week, the comp question. How much should you actually pay a senior AI-capable engineer in 2026. Overpay for pedigree and you torch a year of runway. Underpay for fit and you watch the right candidate sign somewhere else. The point of this issue is to help you set a defensible band so you do neither.
You'll learn why the comp benchmarks you're reading are misleading, why "AI Engineer" is not one job, and how to set a defensible band before your next offer goes out.
Not a subscriber yet? Sign up here
| Why the headline numbers are lying to you
You've probably seen the numbers. Senior AI engineers at $200-312K base. Total comp $300-500K. Levels.fyi median sits at $211K base. OpenAI new grads getting $300K retention bonuses on a 2-year vest. Meta sign-on packages over $100M for elite researchers.
That's the data founders walk into offer conversations with. Almost all of it is wrong for an early-stage startup.
Levels pulls from a self-reported salary dataset that skews heavily Big Tech. Google AI Engineer ranges run $183-583K. OpenAI engineers $249K to $1.28M+. Scale AI $234-642K. When you blend that into a "senior AI engineer median," what you get is a Big Tech anchor, not a market price. The number on screen doesn't apply to a Series A startup hiring its third engineer.
Triangulate Pave's Series A engineer data, Ravio's early-stage band, and YC's published $160-180K base recommendation. The realistic early-stage discount sits closer to 20-30% off the FAANG-anchored numbers. Senior AI-capable engineer at an early-stage US startup, you're looking at roughly $150-185K base and $200-260K total comp. That band covers the two profiles most early-stage startups actually hire, Applied AI and ML Engineer. Research and narrow specialists sit above it. We'll break the profiles down in the next section.
Geography matters here. SF Bay Area or NYC, you're at the top of the band or above. Remote-distributed or non-tier-1 cities, mid-band is realistic. London or Berlin, senior engineers run 30-40% below US comps before currency conversion. Anchor your band to the market you're hiring in, not the global average.
Benchmark off Levels.fyi without applying the discount and you'll either overpay for the wrong profile or panic and lose the right one.
| Five jobs hiding behind one title
Here's the structural problem behind the comp confusion. "AI Engineer" is doing too much work as a job title. There are at least five distinct profiles wearing it, and each one has a different price:
➡️ Applied AI Engineer
Wires up LLM APIs. Builds RAG pipelines. Ships AI features that hit users. Standard senior SWE comp plus a 15-20% premium. You need this if you're shipping AI features users actually touch.
➡️ ML Engineer
Trains and deploys classical models. MLOps. Feature engineering on real datasets. Standard senior SWE comp plus 10-15% premium depending on production scale. You need this if you're training models on your own data, not just calling APIs.
➡️ AI Research Engineer
Fine-tunes foundation models. Builds novel architectures. Reads papers and ships from them. The Big Tech and frontier-lab profile. The $300-500K TC anchor everyone keeps quoting. You probably don't need this. If you genuinely do, you'd already know.
➡️ AI Infrastructure Engineer
Vector databases. GPU orchestration. Inference cost optimisation. 35-45% premium on senior SWE rates where the work justifies it. You need this if vector DBs, GPU costs, or inference latency are your bottleneck.
➡️ SWE who uses AI tools
Not really an AI engineer. Uses Cursor or Claude Code at FAANG. Often labelled as "AI Engineer" on the CV anyway. Standard senior SWE rate. Don't pay them an AI premium just because they use Cursor.
This is a deeper topic than I can do justice to in one section, so I'll come back to it. The point is simple. Founders read "$300K TC for senior AI engineer" and assume it applies to whoever they're hiring. It doesn't. Match the comp to the profile, not to the title.
| The AI premium is narrower than you think
The AI premium is real. It's also smaller than the headlines suggest.
Anchor your senior SWE band first. At early-stage in the US, that runs roughly $150-160K base and $200-220K TC. The premiums in this section sit on top of that anchor.
Ravio's 2025 data shows an 18.7% AI specialist premium in the IC track, up from 15.8% in 2024. Pave's data lands lower. 12% IC track premium, 3% in the management track.
Founders treat the premium as binary. Either you're an "AI engineer" and you get it, or you don't. The data is more granular. The 35-45% premium that gets thrown around online only applies to narrow specialist work. LLM fine-tuning. Production agentic systems at scale. Inference optimisation. If your candidate built an internal tool with the OpenAI API at their last job, that's not the 35-45% bucket. That's the 15-20% bucket.
Anchor your offer to the work the candidate has actually shipped, not the buzzwords on the CV. A candidate who fine-tunes Llama on a real dataset every week is worth the higher premium. A candidate who has "AI" three times on their LinkedIn but spent the last year in meetings is not.
| The $780K mistake
A review of 50 failed AI hires from 2025, published by Fonzi AI and authored by Sammi Cox, had one that stood out.
A Series B startup spent six months recruiting a Senior AI/ML Engineer from a FAANG company. Stanford degree. Recommendation systems experience. They closed at $780K total comp.
Three weeks in, the engineer couldn't implement a basic data pipeline. The last three years of their FAANG career had been spent in meetings, reviewing code, and managing outsourced teams. They hadn't shipped production code in a long time.
Across the 50 failed hires, 28% were skilled engineers who didn't match the work style of a startup. They could code. They could architect. They couldn't sit in a Slack channel at 11pm and unblock themselves on a deploy.
I've seen the same pattern play out at seed with much smaller numbers. A $200K base hire with the right logos who needs three scoping meetings before they can write a line of code. The dollar amounts are different. The trap is identical.
The lesson isn't that FAANG engineers can't ship. It's that pedigree without recent ship velocity is the trap. The Mission and Depth interview from Issue #19 is your filter. Pose a live problem from your roadmap. Watch them work. If they can't sketch the trade-offs in 60 minutes, the offer is dead before it goes out.
| The equity squeeze
Cash is one half. Equity is where the conversation has shifted.
PitchBook reported in late 2025 that ML candidates are now asking for as much as 3x the typical early employee equity grant, putting them close to co-founder territory. Recruiter-side surveys put the figure at over 40% of senior AI specialists now getting more than half their total comp via equity or token grants. The "match base, lean on equity" play that worked two years ago doesn't.
Pave's 2026 founding-engineer data has the baseline grant moving from 0.75% to 1%. Each subsequent engineer is keeping more equity than they used to, with the standard per-hire dilution shrinking from 20% to 10%. The market got tighter, and founders are absorbing the cost in equity rather than cash.
If you're a 5-person team competing with OpenAI for a senior engineer, you cannot beat them on cash. You also cannot beat them on prestige. What you can offer is meaningful equity in something they have a hand in shaping. 1% baseline grant for a founding-engineer-equivalent. Tell them when the next equity grant happens. Even "we'll review at the 18-month mark" beats silence. Transparent dilution math.
If your equity offer requires a calculator the candidate doesn't have access to, you've lost the close before the conversation started.
| What to do this week
Five things to fix before your next senior AI offer.
Throw out your Levels.fyi benchmark: Or apply a 20-30% discount to anything FAANG-anchored. Use Pave's Series A data, Ravio's startup band, and YC's published $160-180K range as your anchors. They're closer to your reality.
Pick the profile before you set the band: Applied AI? ML? Research? Infrastructure? The premium attached to each is different. Setting the band before you've nailed the profile is how you overpay or lowball.
Test ship velocity before you negotiate: A Mission and Depth interview where the candidate works through a problem from your roadmap. If they can't, the comp question is irrelevant.
Set the equity floor at 1% for founding engineers: Tell them when the next grant happens. Make the dilution math transparent. If you can't explain the equity in two sentences, the candidate hears "this isn't real."
Stop trying to match Big Tech cash: You can't. Reframe the offer around autonomy, ownership, speed of progression, and equity upside. Issue #4 covered the one-sentence offer logic. Same principle, applied to comp packaging.
The senior AI engineer market is real. The compensation is real. The trap is treating the entire band as one homogenous number anchored to FAANG datasets. Overpay for pedigree, you burn runway. Underpay for fit, you lose the right candidate. Match the profile to the work, the work to the band, and the band to the offer. The headlines are noise.
Cheers
Neil