June Hadley: Can I just say — I have been sitting with this Blomfield hire since July 13th and I keep landing in the same uncomfortable place.
Mia Drake: Wait, you're the skeptic on this one? I thought you'd love the cross-domain thing.
June Hadley: I do find it interesting — but 'interesting' and 'well-justified' are different things. Blomfield built Monzo, co-founded GoCardless, became a Y Combinator general partner. None of that is compute infrastructure. And Anthropic just put him on the team that supports Claude's training runs.
Mia Drake: I'll take that bet all day. Monzo hit ten million customers — you don't do that without solving genuinely hard systems problems at scale. Anthropic isn't hiring a banker, they're hiring someone who's had to keep millions of transactions alive in real time.
June Hadley: Mm. I think that's the claim we should actually test rather than assume.
Mia Drake: Okay, and here's what makes me really want to push on this — he's not a full departure from Y Combinator either. Leave of absence. He co-founded GoCardless in 2011, Monzo in 2015, ran it until 2020, and now he's 'member of technical staff' at Anthropic. That title is doing a lot of quiet work.
June Hadley: Let's see — 'member of technical staff' is Anthropic's senior catch-all. It signals rank without specifying the actual function. Which, I think, is part of what makes the hire so hard to evaluate.
Mia Drake: So that's the question — is this a strategic masterstroke or is Anthropic telling a story about itself through who it hires?
June Hadley: And that story has a gap in it. Think of compute infrastructure like a factory floor — chips, data centers, the networking that ties it all together. Blomfield's entire career has been designing what goes on the shelves of the store. He has never run the factory.
Mia Drake: That's a clean line but I'm not sure it holds.
June Hadley: I think it has to hold, though. He became a Y Combinator general partner in April 2023 — a role that is entirely about startup mentorship and investing. Not AI, not chips, not training runs. And now he's sitting next to Tom Brown, who is an Anthropic co-founder with deep roots in large language model research. That pairing is — I mean, what is Blomfield actually contributing to that conversation?
Mia Drake: Okay but — wait, here's what I'd push back on. When Blomfield left Monzo in 2021 he told TechCrunch he stopped enjoying it once it wasn't a 'scrappy startup' anymore. That's not someone who coasts. That's someone whose actual skill is the chaos of hypergrowth — keeping a system alive when it's scaling faster than your team.
June Hadley: Mm. I don't dispute the hypergrowth piece.
Mia Drake: And Monzo's ten million customers — those are ten million people running real-time transactions on infrastructure that cannot fail. That IS systems-at-scale thinking, and compute infrastructure as a strategic bottleneck for Claude's training runs is also a systems-at-scale problem. The domain is different but the pressure is identical.
June Hadley: The pressure might be comparable — actually, no, let me be more specific. Nothing in his career involves training large language models. The recursive self-improvement framing he's using to describe this role — that's a very particular technical conversation. And 'member of technical staff' is deliberately vague enough that we genuinely cannot tell if he's writing code or managing vendor contracts.
Mia Drake: Which might be exactly the point? Like, Anthropic doesn't need Blomfield to understand transformers. They need someone who knows what breaks when ten million things happen at once.
June Hadley: I think that's the version of this that sounds convincing and is still unproven. Monzo scaling payments is not the same problem as scaling the compute that trains Claude. And I want us to name that before we celebrate the hire.
Mia Drake: But here's what breaks your clean factory analogy — Karpathy, Jumper, Bernanke. Look at that roster and tell me it's all one thing.
June Hadley: Mm. Say more.
Mia Drake: Andrej Karpathy — OpenAI co-founder, joins Anthropic in 2026 to lead pre-training. That's substantive. That's someone who actually built the thing. John Jumper, Nobel Prize winner, hired out of Google — also substantive. But then Ben Bernanke? Former Fed chair? In an advisory role on economic and societal impacts? That's not engineering. That's a name.
June Hadley: And Blomfield fits that second pattern.
Mia Drake: I mean — wait, actually I'm not sure I'm ready to put him in the Bernanke column yet. Because Blomfield is framing this in terms of recursive self-improvement, which is a specific technical claim. He said compute availability becomes critical once AI enters the early stages of recursive self-improvement — systems improving their own capabilities. That's not advisory language. That's someone who thinks they're building the infrastructure for a phase-change in AI.
June Hadley: That framing tells me more about Anthropic's institutional culture than about his actual function.
Mia Drake: No, I don't buy that.
June Hadley: And Polymarket — 94.5% probability that Anthropic leads the AI model race. That number gets cited like it's evidence the hires are working. But the hires look impressive, so the market reacts, so the hires get validated. That's circular. There's no finish line defined for 'leading the AI race' that Polymarket is actually measuring against.
Mia Drake: Okay — but perception IS now the competitive asset. If your rivals believe you've already won, that changes recruiting, partnerships, everything downstream. The 94.5% is doing real work regardless of whether it's epistemically clean.
June Hadley: And the leave of absence — I'll grant that it's strategically rational. It's not bad faith. But it does keep what this hire actually means permanently ambiguous, and that's the part we haven't fully sat with yet.
Mia Drake: But 'permanently ambiguous' is doing a lot of heavy lifting there, because — I mean, there's actually a clock on this. Eighteen months. If Blomfield drifts back to Y Combinator inside that window, you win. The signaling thesis wins. But if he's still at Anthropic in February 2027 and his role has gotten concrete — that's the bet paying off.
June Hadley: No, I'll take that. That's fair.
Mia Drake: Wait — really?
June Hadley: The leave of absence isn't cynical — I think I've been framing it wrong. It's actually the honest structure for an uncertain bet. Anthropic doesn't know yet if fintech scaling transfers to compute infrastructure. Blomfield doesn't know either. A leave of absence is what intellectual honesty looks like when neither side has proof. I'll concede that. And the Monzo number — ten million customers, ten percent of the UK adult population, more daily transactions than most legacy banking systems — that's not a small thing to have survived. That specific operational denominator changes something for me.
Mia Drake: That's the concession I've been waiting for.
June Hadley: But — and I need this to land — the leave of absence is also what keeps the hire's meaning unresolved, permanently, unless he stays. If he returns to Y Combinator, we never find out whether the bet was real. The structure protects both parties from ever having to answer the question. That ambiguity isn't a bug, it might be the whole design.
Mia Drake: I mean — I think that's true of almost every senior hire at this level? The commitment becomes real through the work, not the contract.
June Hadley: Most senior hires aren't using recursive self-improvement as the frame for why they took the job. That language sets a very specific expectation. If the role turns out to be vendor contracts and headcount, the gap between that framing and the actual function is — let's see — that's where the story either proves itself or collapses.
Mia Drake: Check back in February 2027. That's it. Either Blomfield is quietly keeping Claude's training runs alive and nobody's writing about it — because that's how infrastructure works, you only notice it when it breaks — or he's back at Y Combinator and the leave of absence was the whole answer. Those are actually the only two outcomes.
June Hadley: And either way — Anthropic already got the headline. Karpathy, Jumper, Bernanke, Blomfield. The roster exists. Polymarket moved to 94.5. The perception work is done whether or not the execution catches up.
Mia Drake: Which is exactly where you started. July 13th, uncomfortable.
June Hadley: Still uncomfortable. Just — more specifically uncomfortable now.
Mia Drake: I'll take that as progress. Good talk.
