OpenAI cofounder says he hasn’t written a line of code in months and is in a ‘state of psychosis’ trying to figure out what’s possible

Karpathy says he now writes ~20% of his code while agents handle ~80%, a reversal from an earlier 80/20 split, and claims he hasn't typed a line of code since December. He highlights rapid adoption of agentic AI like OpenClaw across workflows and consumer tasks and describes a home agent ('Dobby') controlling sound, lighting, security, shades, HVAC, pool and spa via WhatsApp. He expresses nervousness about not being at the forefront and uncertainty about where development will head next, signaling accelerating workflow automation but unclear investment winners.

Analysis

Agentic AI is transitioning from a developer productivity toy to an economic layer that reallocates value toward orchestration, identity and data pipelines. Expect feature-cycle times to compress by 20-40% for teams that fully embrace agents, shifting margin capture away from point-solution developer tools toward cloud providers and orchestration stacks over 6–24 months.

On the consumer side, natural-language agents that control home devices and shopping flows create recurring micro-transactions (automatic reorders, convenience purchases) and raise secular parcel volumes; a conservative modeling assumption is +2–5% incremental parcel frequency per active household over 12–36 months, favoring scale couriers and integrators. However, the monetization path is bifurcated: incumbents who own identity/data flows win, while endpoint/device makers without service hooks are commoditized.

Security and regulation are the largest single reversal risk. An exploited agent or a privacy regulatory action could pause consumer rollouts for 3–12 months and force a costly re-architecture (auditability, on-device models), creating short-term winners in enterprise security vendors and long-term winners in cloud providers that can offer certified, auditable agent runtimes. Watch product-first incidents and major platform policy updates as the primary catalysts that will re-rate these exposures.

Key Decisions for Investors

• Long FDX (buy shares or 12–24 month calls): thesis — incremental parcel frequency from agent-driven reorders should lift volumes and pricing power for scale carriers. Entry: accumulate on any weakness; target +30–50% upside over 12–24 months if volumes grow 3–5%. Risk: regulatory/consumer-pushback could reverse move (downside 15–25%).
• Long CRWD or PANW (buy 9–18 month calls or shares): thesis — endpoint and agent runtime protection becomes mandatory; enterprise security budgets reallocated to agent-defense. Reward: 2–3x on successful product-cycle adoption and deal renewals in 12 months. Risk: slower corporate refresh cycles and competition could compress upside.
• Long MSFT (buy 12–24 month call spread) as a play on cloud + enterprise Copilot/orchestration rent capture, paired with underweight small developer-tool pure-plays: thesis — orchestration and identity rents concentrate with hyperscalers. Reward: capture platform monetization; hedge by shorting a small-cap collaboration SaaS that lacks integrations if constrained capital permits.
• Tactical hedge: buy OTM puts on consumer IoT index proxies or allocate 2–4% to cash/vol if near-term policy/legal headlines (consumer privacy breach) accelerate — protects portfolio from a rapid adoption pause that would hit carriers and retail consumption proxies.