Just out of curiosity - Rewriting Bitcoin [ burnsteem100]

Claude's Take

Me: How long would it take for a Claude agent with an unlimited token budget to write the bitcoin code today, if it didn't exist already?

Claude: Fun question. Let me break down what "writing Bitcoin" actually involves versus what an agent could bang out quickly.

What's actually hard about it

The original Bitcoin codebase (Satoshi's 2009 release) was maybe 10-15k lines of C++. Writing code of that size is not the bottleneck for a modern agent - that's an afternoon of generation, even accounting for iteration.

The bottleneck is the design, not the typing:

• Novel cryptographic/economic invention: proof-of-work as Sybil resistance, the longest-chain rule, UTXO model, difficulty retargeting, the halving schedule, script system. This wasn't "known" in 2008 - it required synthesizing ideas from hashcash, b-money, and prior digital cash failures into something that actually solved double-spending without a trusted third party. An LLM in 2026 has all of this in its training data, so it's not inventing anything - it's reproducing a known solution. That's the key asymmetry.
• Getting the incentive design right without being able to simulate it: things like the 10-minute block target, difficulty adjustment period, and fee market weren't obviously correct in advance - they were judgment calls that turned out to work, refined through years of real-world stress testing (block size debates, selfish mining concerns, etc.)
• Networking/consensus edge cases: handling chain reorgs, orphan blocks, timestamp manipulation resistance, malleability - a lot of this got hardened over the following years through actual attacks and bugs found in production, not in the initial design.

So, timeline estimate

If the goal is "produce code functionally equivalent to the 2009 release, knowing what we now know" — a Claude agent could plausibly do this in hours to a couple of days of wall-clock time (mostly bounded by tool-call round trips, compiling, testing edge cases), because it's reconstructing a known, well-documented design rather than inventing one.

If the goal is "invent Bitcoin from a blank slate, the way Satoshi did, without benefit of already knowing the answer" — that's a different question entirely, and not really one where "unlimited tokens" helps much. It's not a compute-bound problem, it's an insight-bound one. Satoshi's contribution wasn't lines of code, it was the specific synthesis of ideas that made the whole system self-consistent. An agent with unlimited tokens but no access to "the answer already exists" would face the same uncertainty any human researcher did — you can't brute-force your way to the right incentive design by generating more tokens, you need something like the multi-year process of proposing, testing, and revising against adversarial conditions.

The honest short answer: code generation, hours; the actual intellectual contribution, not really a "how long" question at all - unlimited tokens mostly just lets you explore a design space faster, it doesn't guarantee you land on a good one.