AI-Native Org Chart: Which Jobs Disappear, Merge, and Emerge?
how AI agents are reshaping headcount, collapsing traditional job boundaries, and creating a new organizational model where smaller teams deliver more
Organization: size, role composition, roles added or eliminated
2.1 How many people, and what is the role composition?
Frontier labs (tier 2). Anthropic does not publish a precise headcount in the study. It says it “has tripled in size in the last year” and takes in 2–4x the revenue per employee that Meta and Google had at the same five-year age (inflation-adjusted, from 10-K comparisons). The disclosed composition is:
- Technical departments are 56% of employees, down from 61% a year earlier. Technical hiring still grew in absolute terms, but sales, finance and recruiting grew faster.
- Engineers to product managers is about 17 to 1, and rising since 2025.
- The share of technical-department staff not actively coding (no commits in 30 days) fell “by almost half” in the year to April 2026, and managers’ commits doubled. In other words, coordinators now code.
- Administrative functions are deliberately thin (“clerical-light”), covered by agent assistants, automated review and outsourced ticket services.
OpenAI’s clearest unit-level datapoint is its “harness engineering” team. It started with 3 engineers driving Codex, now has 7, and produced about 1,500 merged PRs and roughly 1 million lines over five months, about 3.5 PRs per engineer per day, with the code written by agents.
AI-native startups (tier 3, mostly company or press figures):
- Lovable: $100M ARR with 45 full-time employees (July 2025); about $400M ARR with 146 employees (February 2026), roughly $2.7M ARR per head (TechCrunch arithmetic).
- Cursor/Anysphere: reportedly $100M ARR with about 60 staff (early 2025), $1B ARR by November 2025, $2B by February 2026. Headcount estimates for 2025–26 range from about 150 to about 300; Anysphere does not disclose it. SpaceX’s acquisition closed August 14, 2026.
These companies run at $2M–$10M+ of ARR per employee, against roughly $0.5–0.7M at mature SaaS leaders. Caution: some of that gap comes from AI-native firms buying model inference (cost of goods sold) instead of hiring staff, so revenue per employee overstates labor productivity.
Big tech: there is no public team-level role composition. What exists is aggregate: Salesforce support at about 5,000 (from about 9,000), and Uber at about 5,000 engineers on Claude Code.
2.2 Before vs. after: headcount and role changes (big tech focus)
What has verifiably happened at large companies:
- Support consolidation. Salesforce cut support from about 9,000 to about 5,000; Benioff told The Logan Bartlett Show (August 2025), “I’ve reduced it from 9,000 heads to about 5,000, because I need less heads.” Benioff said agents handle about half of conversations, and the company “no longer need[s] to actively backfill support engineer roles.” Headcount was “rebalanced” into sales. A new role appeared: an “omnichannel supervisor” who manages agents and humans together (tier 3; activist-investor pressure is a confounder). Klarna’s assistant handled 80% of chats in 2025, the equivalent of more than 850 FTEs, per its Form 20-F filed February 2026 (tier 2, audited filing). But in May 2025 the CEO said cost-led automation produced “lower quality,” and Klarna rebuilt a human tier.
- Product-role merger. LinkedIn retired its APM program and replaced it with the Associate Product Builder (APB) program, which trains coding, design and product management together. The first cohort is early 2026, the program runs two years, applicants submit a demo instead of a résumé, and the listed base was $126K–$207K. LinkedIn also created a formal “Full Stack Builder” title and career ladder (CPO Tomer Cohen).
- Junior hiring compression. At Anthropic (a lab, but the best-measured case), postings for juniors “have fallen particularly sharply since December 2025.” In an April 2026 survey, about a third of ML managers and a fifth of SWE managers reported hiring less because of AI.
- Performance-system change rather than org-chart change. Meta made “AI-driven impact” a “core expectation” from 2026 (Janelle Gale memo, November 2025). Shopify requires teams to “demonstrate why they cannot get what they want done using AI” before asking for headcount (Lütke, April 2025).
Merged: PM + design + engineering into “builders” (LinkedIn); support tiers into AI front line + human escalation + supervisor (Salesforce, Klarna); writer roles into prompt/eval roles (Anthropic’s product-content team “rotated into agent prompting and began writing model evals”). Reduced: junior SWE and ML hiring, clerical/administrative staff, tier-1 support. Added: agent supervisors, builders, eval writers, agent-platform and FinOps owners (Uber “back to the drawing board” on AI budgeting), and internal redeployment programs (Anthropic: tenured staff were 50% more likely to move departments each month in 2026 than in 2025).
Get Chier Hu’s stories in your inbox
Join Medium for free to get updates from this writer.
My judgment: most big-tech headcount cuts in 2025–26 were framed as AI-driven, but the auditable link between agents and a specific removed role exists in only a few places (Salesforce support, Klarna support). Treat “AI-driven layoffs” claims as tier 3 unless a role-level mechanism is shown.
2.3 Roles common at traditional companies that don’t exist at AI-native startups (startup focus)
Typically absent or near-absent at AI-native startups:
- dedicated QA/test engineers (replaced by agent-written tests, AI code review such as Cursor’s Bugbot, and CI gates)
- APMs and project or program managers (founders and engineers own specs)
- tier-1 support staff (AI front line plus engineers on escalation)
- SDRs (AI outbound plus founder-led sales)
- in-house recruiters at early stages
- most administrative and clerical roles
Share of that work done by agents: no company discloses this. ESTIMATE, with the derivation:
- Anthropic’s time leverage is 6 agent hours per human hour, so agents are 6/7 ≈ 86% of hours. Hours are not value: about 80% of interactive session wall-clock time is the agent waiting.
- Output share of code: Google 75%; Uber about 70% of committed code; OpenAI 99.8% of internal output tokens.
- Shipped-value share: if autonomous agents raise releases by about 30% (NBER), the incremental share of shipped output attributable to agents is about 0.30/1.30 ≈ 23% for a typical adopting developer.
- Estimate: at AI-native startups, agents perform roughly 60–90% of execution tasks in engineering, support tier-1 and outbound drafting, but account for only about 20–40% of incremental shipped value beyond what the same humans would have produced. The rest still depends on human judgment, review and integration. Confidence: low to moderate.
2.4 Boundaries between product, engineering, design and QA; new roles
The boundaries are collapsing toward a “builder” who owns intent-to-production, with specialists retained for taste, architecture and risk. The evidence:
- LinkedIn’s Full Stack Builder ladder and pods.
- Anthropic’s engineer:PM ratio of 17:1, and managers’ commits doubling.
- Non-technical staff running coding-agent sessions: up from 2 per month in August 2025 to 90 per month in May 2026 at Anthropic.
- OpenAI’s report that non-technical users “regularly use Codex to take on coding or technical execution.”
New roles that are actually observed:
- Agent manager/orchestrator (de facto, at Anthropic). 20% of April 2026 survey respondents described their role “shifting from actually doing work to managing Claudes.” One engineer “set up an agent chief of staff to triage agent requests.”
- Harness/agent-platform engineer (OpenAI). Encodes repository knowledge, linters and “doc-gardening” agents so that “human taste is fed back into the system continuously.”
- Eval and prompt writer (Anthropic’s former product-content writers).
- Omnichannel supervisor (Salesforce support).
- AI FinOps owner (implied by Uber and Microsoft spend controls).
- QA is not eliminated; it is moving into “verification engineering”: test harnesses, evals and review automation.
