The modern job is roughly 150 years old. AI is pulling the scaffolding down faster than most people here realize, and the structural question for the valley isn't retraining. It's ownership.
Someone asked me at a coffee shop last week what their kid should study in college. (These conversations have a particular texture now — half question, half braced for the answer.) I said something about computer science and adaptability, the answer I've been giving for three years, and walked out aware I'd given the wrong answer again. The question isn't what to study. The question is whether the entire scaffolding the parent is asking about is still standing.
The job — clocked hours, HR ladders, the career you retire out of — feels permanent because it's the water we swim in. It isn't. As a social institution, the modern job is maybe 150 years old, invented to solve one coordination problem: how to gather thousands of people under one roof and produce at scale. Before that, most people were independent producers — farmers, makers, traders — organizing work around local demand. That older default is reasserting itself. AI is the amplifier.
The large-firm employment model is showing structural fatigue on its own. Administrative overhead, coordination cost, the shrinking half-life of technical skills — in plenty of industries the cost of coordinating people now runs ahead of the value of their output. That would be a slow erosion. AI turns it into something faster.
The benchmark worth watching is DeepSWE. It measures frontier coding agents on original, long-horizon software tasks pulled from active open-source repositories — 113 tasks across TypeScript, Go, Python, JavaScript, and Rust, run in isolated environments with program-based verifiers so the models can't have memorized the answers. It isn't autocomplete. It's whether an agent can land in an unfamiliar codebase, make the change, and not break anything.
The spread between frontier models on DeepSWE runs to 70 points. GPT-5.5 leads near 70 percent. GPT-5.4 sits around 56. Claude Opus 4.7 is at 54. The top of the board is doing real software engineering, end to end. When the unit of production becomes an agentic software loop, organizational scale stops being the moat it was.
The shift already shows up in who's starting things. From 2019 through the first half of 2025, the share of new startups with a single founder rose from 23.7 percent to 36.3 percent. That's structural, not a blip. On the AI-native side, solo founders crossed 2,600 in the first quarter of 2026 — roughly double their level two quarters earlier — while multi-founder startups grew far more slowly.
The minimum viable team is shrinking toward one. Not because going solo is romantic, but because AI now covers the technical, marketing, and operational roles you used to need to hire just to ship.
Once a machine can reason, write, persuade, and optimize, you can't define human value by output anymore. If an agent out-produces you on a task, output is the wrong scoreboard. What's left — what compounds — is judgment, taste, stewardship, and the willingness to take agency in the first place. The near-term winners look like human-plus-agent teams. People deciding what's worth doing. Agents handling scale and repetition. The cultural conversation here has to move from "train for a job" to "define a problem worth solving."
This is where the local story gets misread. The threat in the Coachella Valley isn't mostly the frontline worker. It's the back office. Revenue management, scheduling, marketing, billing, HR — the coordination layer of our service-heavy small businesses is exactly what agents are best at absorbing. That's where the work quietly thins out, and it's why reskilling alone is the wrong answer.
If the value moves to whoever owns the agentic systems, the structural question for the valley is ownership, not just training. A region of people retrained to operate someone else's platform is still a region exporting its margin. Part of why AI Coachella Valley exists is to make the agent-readable infrastructure of this region locally legible and locally owned — not delegated to platforms headquartered somewhere else.
Not a jobs program. Teach problem definition before tool use — point people at a real, repetitive, expensive problem in a real business and help them find the agents that solve it. Stand up an AI-first incubator locally: compute credits, agent templates, legal and finance playbooks, so a founder can ship in weeks. Run short, prize-driven hackathons with public demos, because momentum compounds when people watch their neighbors ship. Pair experienced operators who understand workflows with builders who write code. And build the transition honestly — portable benefits, dignified exits — because if this gets treated as purely a tech story, the region loses the standing to do it well.
None of this is clean. Centralized control of frontier AI could concentrate power in the exact way that kills the agency we're trying to spread. Surveillance, economic coercion, the hollowing of public institutions — real failure modes. Civic and religious institutions are right to be asking hard questions about human dignity here. Any regional plan has to build governance in at software speed — audits, sandboxes, guardrails — not bolt it on after.
The rise and fall of the job is a reinvention, not an apocalypse. Work will exist. It always has. The narrower question for the Coachella Valley is whether we export talent to other people's platforms or own the ecosystem we build here.
For any operator reading this: what's the back-office function in your business that an agent could absorb tomorrow, and who captures the value when it does — you, or the platform that sold you the agent?