Articles

Two premium streaming shows, two writers' rooms, two opposite decisions about AI tools — and the answer to which show survived is more complicated than either side in the ongoing debate about creative labor and artificial intelligence will comfortably admit.

A small delivery operation I help. Route planning took 90 minutes every morning. I wired Gemini to their ops sheet and the job runs in 20 minutes with fewer errors. Here is the prompt architecture.

Generative design tools let architects iterate buildings in minutes, but on a construction site with active crews, the efficiency that accrues to the design office arrives in the superintendent's morning as coordination pressure, safety risk, and uncompensated management work.

Federal agencies spent five years running AI pilots that rarely converted to funded, baseline programs — a pattern the GAO documented exhaustively and the Office of Management and Budget acknowledged as a structural failure of acquisition and governance design.

During a red-team exercise against a banking agent with read and write permissions to customer accounts, an indirect prompt injection delivered via an email summary tool achieved an unauthorized fund transfer in four interaction steps. The agent's prompt hardening had been designed to resist direct injection; the permissions architecture had not been designed at all.

My niece kept getting stuck on quadratics. I built a Socratic-method tutor that never just gives the answer. Her grade went from C+ to A-. Here is what I learned about edtech from watching a 14-year-old use it.

The Big Four accounting and advisory firms collectively spent an estimated $2.8 billion on AI infrastructure in 2025, subsidizing AI tool costs for clients while defending audit and advisory fee levels. In 2026, that subsidy is unwinding: the firms are beginning to charge for AI-assisted work at rates that reflect the efficiency gains, not at rates that conceal them.

I had three ways to get structured output from an LLM. I had actual production data to test against. I benchmarked all three on reliability, latency, and prompt overhead. The winner was the one I did not expect.

A health-tech team shipped an AI clinical-note summarizer with a plaintext prompt exposed directly to clinicians; daily active use was 8% of target at four weeks post-launch, and qualitative feedback indicated that clinicians did not understand what the system would accept, what it would produce, or what to do when it failed.

The last utility meter reader in her district retired in February, and the job that nominally replaced hers pays 22% less in inflation-adjusted terms — a gap that maps the distance between what 'workforce transition' sounds like from a corporate communications office and what it looks like from the ground.

I let an agent handle Tier-1 support for 7 days. It resolved 64% of tickets without escalation. It also almost promised a refund it could not actually issue. Here is what held up and what did not.

I rebuilt a workflow I had been running manually for six months as a three-agent LangGraph system. Two of the three agents worked on the first try. The third one is a story. Here is the build log.