Attimet's 'Prime Radiant' job posting shows the agent-infra gold rush is just beginning

The most revealing artifact in AI hiring right now isn’t a model release. It’s a job posting from Attimet, a three-person YC F24 startup, looking for a “Member of Technical Staff | Engineering & Research” at $125K-$350K with 0.10%-2.00% equity.

The posting is remarkable for what it says about the state of the AI industry, and for what it doesn’t say. Attimet’s stated mission, in full: “Step 1: Build the Prime Radiant. Step 2: Apply it to Financial Markets. Step 3: Manage the world’s assets.”

The “Prime Radiant” is a reference to Isaac Asimov’s Foundation series: a device that holds the entire Encyclopedia Galactica and can project any part of it. For a three-person startup, the ambition is either wildly arrogant or precisely scoped. The job posting suggests the founders believe they can build something like it: an AI system that “expands how quickly a small research team can understand, build, and experiment.”

What’s actually new here is the category, not the company. Attimet is a research lab whose product is the infrastructure for research itself. The founders, Kirthi Banothu (CEO) and Xiaoyu Li (CIO), come from Optiver, DRW, and Argo AI, per the posting. They’re not building a model. They’re building the harness around models: “tools, context, memory, evals, orchestration, observability, and execution environments.”

That list is the entire stack of what the AI industry now calls “agent infrastructure.” And it’s the same stack that every serious lab, from OpenAI to Anthropic to the hedge funds, is quietly building internally. Attimet’s bet is that a small, fast team can build it better than the incumbents, and then point it at financial markets.

The real signal: what “research lab with a real-time feedback loop” means

The posting’s first bullet under “Why you should join us” is the one that matters: “We’re building a research lab with a real-time feedback loop.”

That is a precise technical claim, not a vibe. A real-time feedback loop means the system learns from every experiment, every trade, every failed eval, and feeds that learning back into the next iteration. For a three-person team, that’s the only way to compete with labs that have hundreds of researchers. It’s also the same mechanism that quantitative trading firms like Optiver and DRW have used for decades: form a hypothesis, test it in milliseconds, keep what works, discard what doesn’t.

The founders are importing that culture into AI research. The posting says the edge is “the speed at which we can form ideas, build systems, run experiments, and learn from reality.” That’s not a mission statement. That’s a description of a quantitative trading desk applied to LLMs.

The first 30 days section makes the tempo explicit: “By 15 days: take ownership of a meaningful technical problem end-to-end. By 30 days: we want you independently identifying what matters and driving projects that materially increase what the team can do.”

Most AI research labs cannot honestly make that promise. The hiring manager at a frontier lab would laugh at the idea that a new grad owns a meaningful technical problem by day 15. Attimet is betting that the bottleneck in AI research is no longer compute or model quality, but the speed of the human-in-the-loop iteration cycle.

The compensation is the tell

The range of $125K-$350K with 0.10%-2.00% equity is wide, and that’s deliberate. At the low end, Attimet is paying like a seed-stage startup. At the high end, it’s paying like a hedge fund. The equity range is equally telling: 0.10% is what you’d offer a senior engineer at a company that might be worth a billion dollars. 2.00% is what you’d offer a co-founder.

The posting says “Any (new grads ok)” and “You care more about what works in practice than following established patterns.” It also says “We do not do LeetCode-style interviews.” That’s a direct appeal to the person who has built agent harnesses in their spare time, not the person who has memorized dynamic programming.

The visa restriction (“US citizen/visa only”) is worth noting. For a company that wants to “manage the world’s assets,” restricting the talent pool to US citizens is a constraint, not a choice. It suggests the founders are either cautious about the regulatory environment or they’ve decided the speed of hiring matters more than the breadth of the pool.

What this means for the AI industry

The most important implication is that agent infrastructure is becoming a standalone category, with its own hiring market, its own compensation bands, and its own career path. Attimet is hiring for a role that did not exist two years ago: someone who builds “LLM-powered systems and agent harnesses that help us research, engineer, and operate faster.”

That role is the definition of a meta-tool. It’s building the tools that build the tools. And it’s the same role that every AI-native company, from the trading desks to the research labs to the vertical SaaS startups, is trying to fill right now.

The second implication is about capital allocation. Attimet is a three-person team with a $125K-$350K budget per hire. That’s a tiny amount of money compared to the training runs at frontier labs, but it’s a massive amount of leverage if the harness works. The founders are betting that the marginal dollar spent on agent infrastructure returns more than the marginal dollar spent on compute.

The third implication is about the “Prime Radiant” itself. If a small team can build a system that lets a few researchers do the work of a hundred, the economics of AI research change fundamentally. The bottleneck shifts from headcount to the quality of the feedback loop.

The financial markets angle is the least surprising part. Trading firms have always been the first to adopt new technology that gives them an edge, and LLMs are no exception. What’s new is the explicit claim that the same infrastructure that accelerates research can also manage assets. That’s a bold claim, and the posting doesn’t pretend to have solved it. It’s a three-step plan, and the company is still on step one.

The question nobody is asking

The posting raises a question that nobody in the AI industry is asking out loud: what happens when the research lab itself becomes the product? Attimet is building infrastructure to accelerate its own research, and then plans to apply it to financial markets. But the infrastructure it’s building is exactly what every other lab needs.

If Attimet succeeds at step one, it will have a choice: keep the infrastructure proprietary and use it to trade, or sell it to every other lab and research group that wants the same speed. The “manage the world’s assets” line suggests they’ve chosen the former. But the posting’s emphasis on “evals, orchestration, observability” suggests they’re building something that could be productized.

The founders’ backgrounds at Optiver, DRW, and Argo AI suggest they know the value of a proprietary edge. But they also know that infrastructure is a business, and that the market for agent harnesses is growing faster than the market for any single trading strategy.

For AI builders, the takeaway is direct: the agent-infrastructure layer is where the leverage is right now. The models are commoditizing. The harness around them is not. Attimet’s posting is one of the clearest articulations of that thesis, and it’s hiring for exactly the skills that the next wave of AI companies will need.

The Prime Radiant is a fictional device. But the feedback loop it represents, the ability to compress the cycle from idea to experiment to learning, is real, and it’s the competitive advantage that every AI company is now chasing. Attimet is just the first to say it out loud in a job posting.