“Not lightly was the fire delivered, Nor gladly the doom foreseen; For wisdom looks beyond the hour, And weighs the cost unseen.”
- Robert Bridges’s Prometheus the Firegiver
Last week we talked about the SaaS dislocation. But running quietly alongside that has been something that might matter even more: just how fast the frontier models are moving, and what it actually feels like to use Claude or OpenAI in real work.
For most of my career, when markets dislocated around an event, my playbook barely changed. I’d look for historical analogs, stress-test whether the pattern still applied, and then hunt for mispricings where conviction could become position sizing. The real cost was never capital. It was time. Doing it properly meant donating seven or eight hours of a weekend to reading reports, cleaning data, building valuation models, and iterating through scenarios until the picture stopped wobbling.
When I started using frontier models about a year ago, the promise was obvious, but the experience was… messy. Hallucinations were routine. The reasoning drifted. And the output often needed so much correction that you weren’t sure you’d saved anything. That’s changed materially. Accuracy is up, the logic holds together more often, and the big shift is cycle time. Work that used to swallow a weekend now fits into a few hours, sometimes less.
That leads to the core point: two demand engines are compounding at the same time — efficiency and expansion.
The efficiency effect is the simple one. Same workflows, faster and cheaper. A multi-scenario model that used to take an afternoon can show up in minutes. Even if nothing else changes, that’s real ROI: you’re compressing the cost of doing today’s work.
The expansion effect is the one people underestimate — and it’s the bigger deal. When analysis gets cheap enough, you start doing work you never would have justified before. Backcasting thirty public companies with DCFs and sensitivity grids was always technically possible, but practically uneconomic. Now it’s an afternoon. That isn’t substitution. It’s creation. “I wonder if…” stops being a thought and becomes a spreadsheet, a memo, a view. And that’s where most market-sizing arguments miss the plot. They count the old work done faster. The real market includes all the work that never happened because time and effort made it irrational.
This is also where frontier agents start to diverge from the older idea of “algos.” Traditional trading systems live inside tight instructions. They’re bounded, reactive, and mostly feed on price and a narrow set of signals. Frontier agents are closer to junior operators with superpowers. Give them a goal and they can pull data from filings and transcripts, cross-check it against news flow, generate variants of an investment memo, draft outreach to industry experts, and tee up the next set of questions — all while you’re still deciding what the right frame is. They don’t just watch the tape. They move around the information environment.
So what’s actually happening?
At one level, finance is getting a straight-up time compression shock. The job has always been: take too much information, turn it into a decision, and do it before everyone else. As data volume and velocity keep rising, these models make that translation faster. Not in a magical “perfect prediction” way — more in a very practical way: they help you cover more ground, run more iterations, and tighten the feedback loop between a hypothesis and the evidence.
The strategic upside improves too. You’re not only faster; you’re wider. The models can surface patterns and second-order interactions that are easy to miss when you’re juggling ten threads in your head and a hundred tabs in your browser. And there’s a quiet power shift inside that. It rewards genuine domain expertise — not because expertise is suddenly new, but because the old bottlenecks around execution are disappearing. Time, iteration bandwidth, and dependence on layers of junior throughput stop being the constraint.
Humans are still very much in the loop. But the loop changes shape. If you’re optimizing for “accurate enough to make a good call” rather than courtroom-grade precision, frontier models get you to a strong first draft insanely fast. The scarce inputs move up the stack: judgment, framing, interpretation. Not spreadsheet stamina.
That’s why it feels Promethean. These tools are broad, surprisingly general, and getting better in a straight line. They democratize work that used to require serious time, capital, and labor — and in doing so, they erode moats built on process capacity rather than real insight.
One counterweight is worth stating explicitly. None of this speed and breadth automatically translates into better outcomes, and in some cases it can make things worse. When everyone gets access to the same cheap analytical horsepower, errors can synchronize, narratives can spread faster, and the system can become more crowded and more fragile, not less. More scenarios can mean more overfitting, more false confidence, and more activity that looks like insight but is really just noise produced at scale. And the advantages won’t disappear evenly: proprietary data, governance, compliance constraints, and genuine judgment still separate people who have the tools from people who can actually use them well. The ceiling rises, but so does the speed at which bad frames and shared mistakes can compound.
Which sets up the obvious thought experiment: if more people have these tools, markets should get more efficient. Mispricings should get spotted faster, priced faster, and competed away faster. If that’s true, where does alpha migrate?
Probably toward the parts of the world that refuse to be cleanly computed: regime shifts, geopolitics, policy decisions, regulation, structural flows — and the messy reflex where stories move prices and prices reshape stories. In other words, interpretation under uncertainty, not calculation under neat assumptions.
There’s a clear upside to that world too. Financial guidance becomes more available — always on, lower cost, and genuinely confidence-building for people who were previously shut out. Over time, that narrows the gap between professionals and non-professionals, and it forces incumbents to compete on judgment and trust, not access and mystique.