Start researching this function and you’ll meet numbers like these. “Senior AI PM total comp $320K–520K.” “AI PM postings nearly tripled in two years.” They circulate separately, and both wobble the moment you trace them. The first sits in a recruiting firm’s salary guide with no sample and no method; the second appears in content-site posts with no source at all.

Here’s what measurement looks like: a tally of salaries actually posted in 12,397 US job ads (Axial Search, January–June 2026) puts the median at $194K, a long way below the claim. But the units differ: $320–520K is total compensation, equity included; $194K is the median of posted salary bands. Compare total comp against posted salary and much of the gap is just whether you count equity. Compare posted bands with posted bands and AI labs still stand apart: one Anthropic PM posting lists $305,000–460,000, about 1.6x the market’s posted median. That’s one lab posting’s band, though, not a statistic about “the floor at labs.”

No corner of this series shows better why every number carries its source and its unit.

One name, two directions

Two live postings, side by side, make it visible.

Build-side (AI PM): Anthropic, PM for Claude Code model performance, $305–460K. Owns the product’s evaluation roadmap and model launch planning, and the requirements include having personally built agentic eval suites (SWE-bench-style). The evals from the Application piece are, here, a hiring criterion.

Sell-side (Solutions Architect): Anthropic, Applied AI Architect, $240–315K. Works alongside sales, product, and engineering as the customer’s technical advisor from discovery through deployment, with requirements of eight-plus years in roles like Field CTO, solutions architect, sales engineer, or product manager, plus hands-on building experience.

Same company, same function, different directions: build-side answers for its own product’s metrics, sell-side answers for the customer’s deployment. Moving between them isn’t hard, though; that architect posting lists PM experience as qualifying, and the market treats the two directions as neighbouring careers.

Sell-side is less covered

Job-hunt content talks mostly about the PM side. Pay runs in two directions, though. Inside the same lab, the build side earns more: in the two postings above, the PM band starts at $305K and the architect band at $240K. Where sell-side pay has climbed is outside the labs. What AI vendors pay solutions architects has moved to a different tier from classic enterprise software. On levels.fyi, OpenAI solution architects show a median total comp of $418K and Databricks $353K (checked 3 September 2026). The caution that matters isn’t self-reporting; levels.fyi verifies against offer letters. It’s the sample. Neither page discloses sample size, and for a niche role like this a median can rest on a handful of entries.

The fast-growing title lives on this side too. Salesforce hires Agentforce Forward Deployed Engineers continuously in the US, and Australian postings come and go. But FDE bands vary widely by company and level (one Salesforce all-levels US posting runs $88,970–287,910 base, up to $316,750 in the big metros), so read the company and the level, not the title. What this direction wants is someone who can both build and explain what they built in front of customers. Few people do both, which is why it pays.

A warning about the titles

LinkedIn’s 2026 fastest-growing US jobs list has no PM and no solutions architect in its top 25; #1 is AI Engineer, #2 AI Consultant. Growth-rate lists structurally favour small bases, so absence isn’t decline. But the list does show where new demand points. And there’s precedent: Airbnb retired the traditional PM title in 2023 and merged product management with product marketing into one group. When every product is an AI product, the “AI” in “AI PM” may simply evaporate.

What the first piece said applies with full force here: prepare for the work the role does, not the title. Proving a product’s value in numbers, and landing deployments that survive at a customer, outlast whatever the title becomes.

The Australian numbers

Permanent solutions architects in Sydney run AUD $165–220k (Morgan McKinley $165–220k, Robert Half $180–213k, 2026 guides). For contracts, the mainstream benchmark puts roles like LLM architect and agent architect at $1,430–1,490 a day (Talent International, March 2026). Figures like $2,000–3,500 a day or a “20–35% AI premium” come from a single specialist recruiter, so don’t set expectations there. In the mainstream guides, the AI premium shows up only at the top of the band, if at all.

The evidence differs by direction

  • Build-side: metric movement attached to something you shipped (conversion, cost, retention), plus a record of running an eval roadmap.
  • Sell-side: the fact that what you deployed at someone else’s shop is still in production. Not a demo; a deployment that’s alive three months later.

The two Anthropic postings above are senior roles. The junior path opens lower on the sell side: vendor solution-engineer and FDE bands come first, and a side project is evidence that works there. If you have one, treat it like a product. Get ten people using it and measure their usage: build-side evidence. Deploy it for another team or company and keep it running: sell-side evidence. The title comes after.

That completes all five functions: what each one measures and what it has to show. From here, the series moves to guides: actually producing each function’s receipts.