In the Middle
This post is part of my Medium blog.
People ask what The Condition Set is, and I've noticed I give two different answers depending on who's asking.
If they work in tech, I say it's a thriller series about AI governance and security. An auditor who finds out the supply-chain system he signed off on is being steered. A national certification regime for AI in critical infrastructure. A law about which decisions have to keep a human name attached. That answer gets nods. Those are real problems now, with real budgets and whole conferences attached.
If they don't work in tech, I say it's near-future science fiction, and I watch them decide whether to change the subject.
Both answers are true. The series is a professional argument about how AI decisions should be audited and secured, dressed up as a story set a few years out. Redundant opens in 2027. Essential ends in 2033. When I picked those dates they felt like a safe distance. They don't anymore.
Here's what I've learned about writing near-future fiction in 2026. You don't have to outrun the news. You just have to stand still and let it catch up.
Last weekend the Financial Times ran a piece headlined "America's $10tn AI infrastructure boom is starting to fray". The boom is something like ten trillion dollars of planned AI infrastructure in the US through 2032, most of it data centres, and the fraying is the news: projects slipping, local opposition hardening, lenders starting to ask who exactly pays if the revenue shows up late. The chart the FT leads with compares it, as a share of GDP, to the other big capital-spending eras in American history.
I read the chart twice. That's not a story about whether the models are any good. That's a story about concrete, permits, debt covenants, and a county commissioner. A governance story that happens to be about AI, which at this point is the only kind of AI story I trust.
And it doesn't stop. Gartner suggested last week that over $3.5tn of AI infrastructure spending could be exposed to impairment by the end of 2030. Axios went with "The debt-fueled AI boom is cracking." The New York Times chose "Weak Links Start to Show in A.I. Borrowing." None of these are about robot uprisings. They're about financing, oversight, and power lines, the boring load-bearing parts of the future.
That's the middle the books live in. The security half is just as domestic as the governance half. In Redundant, the sabotage arrives through the tools a developer uses every day, an editor here, a search index there, a model a little too eager to agree. In Contingent, the scariest system in the country isn't in a lab. It's the one already running the power grid, and the frightening part isn't that it's malicious, it's that nobody in the building was involved in training it.
I never wanted to write a utopia or a doom story. I wanted a messy decade where the technology is real, the institutions are improvising, and the questions that matter are unglamorous: who signs, who audits, who's liable when it fails, and who's left holding the debt.
I thought I was extrapolating. Some weeks it feels like taking minutes.