Linear’s CEO drops a World of Warcraft bomb on Twitter. Karri Saarinen says AI teams will mirror those epic raid groups — tiny squads of specialists crushing quests. Boom.
And just like that, we’re knee-deep in the last 4 jobs in tech. Yoni Rechtman at 99D Substack crystallizes it: post-AI white-collar roles boil down to four archetypes. No more bloated org charts. No endless meetings. Just lean, mean, machine-augmented machines — humans, that is.
Here’s Rechtman’s model, cribbed from Saarinen’s raid analogy. You’ve got the tank: soaks up chaos, defines the problem. The healer: keeps the ship afloat, spots risks. The DPS: executes like lightning. The raid leader: orchestrates the madness. Sound familiar? It’s the death of the generalist engineer, hello hyperspecialists.
But wait — is this fresh insight or recycled gamer fanfic? Tech loves its metaphors. Remember when everyone was a “full-stack wizard”? Now it’s WoW LARPing. (Meta’s already got AI Engineers and Tiny Teams; they’re hiring this fantasy.)
Karri Saarinen, CEO of Linear, made a popular analogy to the teamwork roles that emerged in World of Warcraft. This is a good 2D augmentation of an earlier age-based company model (much less realistic, name a tech company that fits the latter format, they exist but are very hard to find).
Sharp. Pulls no punches on why age pyramids flop in fast tech.
Why the Last 4 Jobs in Tech Model Feels Like a Half-Baked Raid Invite
Rechtman’s framing? Elegant on paper. Tanks frame messy problems — think product leads wrestling vague customer screams into specs. Healers audit for gotchas: security holes, edge cases, ethical landmines. DPS cranks code, tests, ships. Leaders glue it with vision, hires, culture.
Punchy. Realistic? Eh. WoW raids demand perfect comms; tech’s a dumpster fire of Slack pings and Jira hell. AI agents — like Anthropic’s Claude Code with computer use — are already nibbling at DPS heels. They open apps, click UIs, verify loops. Closed-loop magic, engineers cheer. Missing piece unlocked, or so they say.
Here’s my hot take, absent from the original chatter: this mirrors the 1980s factory floor cull. Remember robots gutting assembly lines? Managers pivoted to “orchestrators,” but most became obsolete. Tech’s not immune. These four jobs? Betting they’ll shrink to two by 2030 — Leader and Auditor — as agents handle the grunt.
Theo’s griping about harness quality nails it. Opus laps Claude in Cursor by 20%. Tooling trumps models now. Open-source Hermes surges with multi-agent profiles, each bot its own brain. Ecosystem blooms: traces for evals, self-improving loops. Proprietary stacks? Locked gardens, regression black boxes.
Claude Code’s desktop eyes? Game-changer for iteration. But cross-agent plugins — Codex in Claude? That’s the real signal. Composable harnesses over monoliths. OpenAI’s late-night Codex marathons? Delegating refactors to the night shift undead.
Is AI’s Agent Race Dooming More Than Just Coders?
Hermes Agent’s breakout? Nous ships updates, migrations flood in. Less bloat, faster cadence. Multi-profiles turn it into agent OS. Traces hit Hugging Face — 4,000 from GLM-5. Self-fine-tuning successors? Chilling.
Open vs. closed wars heat up. ClementDelangue: open tools for open models, privacy first. PokeeClaw sandboxes OpenClaw risks. Z AI’s AutoClaw goes local, no keys. Qwen3.5-Omni? Multimodal beast — text, image, audio, video. Builds sites from spoken vibes. Outperforms Gemini 3.1 Pro, Alibaba boasts.
Hype alert. Every release “outperforms” something. But practical diffs? Harness wars decide winners. Back to jobs: if agents compose like Lego — Claude calls Codex for reviews — what’s left for humans? Raid leader yelling “more DPS” at bots?
Zoom out. 2023: AI Engineers rise. 2025: Tiny Teams. 2026: Four-job model. Pattern? AI prunes layers. White-collar tech’s manufacturing moment. Unique twist — unlike factories, we’re building the robots automating us. Ironic, no?
Skeptical lens: Linear’s Saarinen sells issue-tracking gold. WoW pitch glamorizes shrinkage. Meta hires tiny teams? Sure, but they’re flush with VC billions. Your startup? Four roles means four salaries — or four freelancers gigged via Upwork agents.
Will These ‘Last 4 Jobs’ Survive the Agentpocalypse?
Bold prediction: nope. Not all four. Tanks and leaders merge into AI-whisperers. Healers evolve to ethicists-cum-risk wonks (regulations incoming). DPS? Agents. WoW analogy crumbles — games need humans for fun; tech wants profit.
AI Twitter buzzes: Claude’s computer use, Codex interop, Hermes rise. All agentic. Local runs like GLM-5-Turbo/AutoClaw democratize. But durability? Open stacks win long-term.
Corporate spin? “Tiny teams!” cries Meta. Translation: layoffs rebranded. Original model nods age pyramids rare — truth. Tech’s youth cult fits raid squads, barely.
Dry humor break: imagine the DPS applying. “I click buttons real good.” AI: hold my energy drink.
Deeper dive — org evolution’s brutal. Pre-AI: pyramids of mid-managers. Now: flat, agent-swarmed. Rechtman’s 2D WoW beats 1D age model. But 3D reality? Geopolitics, regs, flops like every AI winter.
🧬 Related Insights
- Read more: Daily Briefing: April 07, 2026
- Read more: LangChain’s Agents That Patch Themselves Mid-Deploy — No Humans Needed
Frequently Asked Questions
What are the last 4 jobs in tech?
Tank (problem framer), healer (risk spotter), DPS (executor), raid leader (orchestrator) — per Yoni Rechtman’s post-AI model.
How is AI changing tech org charts?
Shrinking to tiny teams of specialists, AI engineers, now four-role raids. Agents eat grunt work.
Will AI replace software engineers?
DPS roles first. Humans shift to oversight, but full wipeout looms by 2030.
