Where GTM work is going.

Why I expect this

The seats that absorbed Wave 1 compress next. As agent fleets spread, the scarce hire becomes whoever makes them actually produce — the constraint won't be tools, it will be org design.

What crossing takes

Run a fleet: own an agent system's output end to end, with numbers.

What would change my mind

If by end of 2028, postings for GTM-engineering and agent-operations roles at AI-native startups aren't outnumbering SDR postings, this call was wrong.

Why I expect this

As buying agents proliferate, making your company legible to machines — machine-readable pricing, agent-ready product data, citation authority — becomes a funded function. Selling to machines becomes a job.

What crossing takes

Ship one agent-legible surface for a real company and measure what changed.

What would change my mind

If by 2028 no funded companies are posting agent-facing GTM or machine-readable-sales roles, kill this card.

Why I expect this

Referral networks hold the executive layer, where trust is the product. The volume layer shifts to whoever can verify capability — matching proof portfolios to problems, not titles to titles.

What crossing takes

A portfolio of shipped work a recruiter can check in five minutes.

What would change my mind

If title-based postings still dominate hiring volume through 2028 and proof-portfolio screening stays niche, this was my bias, not a trend.

Why I expect this

The seats that stay human are the ones buyers already reserve for humans: accountability, trust, and system design. Klarna's reversal was the early sketch of this boundary.

What crossing takes

A track record of owning outcomes a company would not hand to a machine.

What would change my mind

If AE headcount at AI-native companies shrinks 2026–28 at the rate SDR headcount did in 2024–26, the human-seat thesis fails.

Why I expect this

Every AI-adopting company — in every industry, not just tech — needs someone who rebuilds real workflows around the tools. The Klarna pattern generalizes: the routine layer automates, and a human implementation seat opens upstream of it. The early count already exists: LinkedIn's 2026 Labor Market Report includes forward-deployed engineers among 1.3 million AI-related openings created in two years.

What crossing takes

Owned implementations with before/after numbers, at companies that aren't software companies.

What would change my mind

If by 2028 AI-implementation and forward-deployed roles aren't a standard posting category at non-tech companies adopting AI, this generalization failed.

Why I expect this

Stanford's 2026 Index has agents completing 66% of real-world computer tasks, within six points of human performance. Fleets of near-human agents create a new seat: the person who runs them — QA, escalation, throughput, cost — across support, sales, and back office at once.

What crossing takes

Run an agent system in production and own its numbers.

What would change my mind

If agent-operations / AI-operations titles aren't appearing across support, sales, AND back-office postings by 2028, the seat stayed hypothetical.