Code mode is supposed to be the faster way for agents to call MCP tools. Fewer round trips, smaller context. We raced it against classic tool calling on the same task. Code mode lost by 39 seconds. The reason isn't a code mode problem. Most MCP servers ship input schemas and nothing else, so the agent knows exactly what to send and has to guess what comes back. It guessed wrong three times, and every wrong guess is a full extra turn. So we built a shape cache that remembers what each tool returns and carries it across sessions. The Rematch: - 27% faster than classic, - 47% cheaper - 42% less context. Read how we turned the loss into a win:

Something I routinely am in awe about is the compute amplification every commit has today vs. 20 years ago, at least personally. Back then, my personal computer was only non-idle when I was working locally. I ran tests locally. When I pushed, there was no CI. The tests themselves were minimal maturity (simple unit tests, some e2e). Today when I push, CI exists. It fans out to 200+ jobs across macOS, Linux, and Windows that take ~20 minutes to complete in totality. A successful CI triggers ~8 beefy machines for tip releases. Tip releases trigger Apple notarization machines. Simultaneously, I have dedicated machines running 24/7 doing fuzz testing at basically 100% CPU constantly. Locally, my computer is using CPU much more of the day due to background agents. My average CPU utilization on my M4 Max overnight the last 7 days was over 20%. "Low" but before my average would've been 0% (deep sleep)! Plus, those agents are commanding a ton of GPU utilization elsewhere. I'm not making any judgement here of whether this is good or bad, because I think there are both good elements and bad elements. Simply that its quite astonishing how much compute my daily work amplifies in 2026 vs. 2006.

Outages absolutely suck. I deeply empathize with the pain this is causing our customers, and we’re sorry for the disruption. Some context: we had a highly unusual infrastructure incident at one of our data centers where the cooling systems failed along with their redundancies. Early signs point to storms in Phoenix causing power interruptions at the facility, which took both the primary and backup cooling systems offline. In the meantime, we’ve load balanced across all our other regions to minimize disruptions. Cooling is coming back online, temperatures are starting to stabilize, and we’re bringing more machines back up. We expect full recovery soon. We’re sharing updates in real time on our status page We’ll also share a detail post-mortem.

We raised a $45M Series B led by Peak XV in March, valuing @useblacksmith at $550M. We somehow never got around to announcing it until now; this year has been pretty busy. One thing I’ve been thinking about is how we were both right and wrong about what was coming. When we raised our Series A in April 2025, we believed AI was going to lead to a lot more code being written, and that all of that code would still need to be built, tested, and validated before it could be merged. That part was right, but what we underestimated was how fast it would happen. I have a suspicion that something changed over the holidays. Opus 4.5 came out in November; a lot of developers had time to really use Claude Code over the holidays. I think a lot of people who were still skeptical went back to work in January having fully embraced it. Since the start of the year, the number of CI jobs running on Blacksmith has grown more than 7x. We see customers adopt coding agents, produce way more PRs, and then run into the next problem: how do you actually validate all of this code and get it merged? There are a lot of people working on making code easier to produce. What I find strange is how little attention goes into everything that has to happen after the code is written and before it can be merged. The labs and hyperscalers are mostly focused elsewhere, yet developers and agents spend an enormous fraction of their time here. A lot of companies touch parts of this problem. For us, it’s *the* problem. We started Blacksmith by building a much better CI cloud, but ultimately what we care about is helping developers and agents validate and merge code faster. I sometimes tell candidates that if we don’t solve the whole thing, I’m not sure who will. I don’t think we fully appreciate yet how much more software is going to be produced. The infrastructure we have today wasn’t designed for that volume. There’s a lot we need to rethink from first principles, and I think we’re barely scratching the surface. If you’re extremely good at what you do and this is a problem you want to spend the next several years working on, we’re hiring in SF and NYC:

Peak XV (cc: @sjs_day1, @noahpbradford) is leading a $45M Series B round in Blacksmith (YC w24) - the leading platform for code infrastructure - alongside @snowmaker at YC and @ejn at Google Ventures. While it’s 10x faster and easier to write code, the infrastructure required to validate, test, secure, and deploy that code hasn’t kept up. Blacksmith is changing that. They sit at the intersection of three large markets: infrastructure, compute, and tokens - they’re growing VERY fast. Long thread: (1) AI code generation could be the largest software market ever: Spend has already crossed $50B ARR in the 3 years. All software is ultimately an abstraction of code. In 10 years, it’s possible token spend could still 100x as long-running background agents become capable of working autonomously for hours, days, and weeks and eventually one-shot entire applications. (2) The AI coding market will likely be fragmented: We are not intelligence-constrained for every token. We are seeing capable open-source models that are faster and cheaper for many tasks. Today, developers already use multiple coding harnesses and models simultaneously. (3) A massive new code execution, validation and infrastructure market is emerging. In a fragmented-model world, where we can write code 10x faster, the bottleneck has shifted from writing code to infrastructure and everything that happens after code is written. We saw a similar pattern in cloud and SaaS: observability, security and monitoring became huge downstream markets. There will be a similar wave in AI code, but likely bigger. As companies run agent swarms, long-running background agents across different model providers, the volume of code that needs to be executed, secured, tested, and validated will increase dramatically. Importantly, the companies that win this market will need to be agent-neutral and model-agnostic. (4) Blacksmith has the potential to become a defining developer infrastructure company. They already power the critical infrastructure for thousands of frontier, AI companies, including Exa, Serval, Reflection, Supabase, Mercury, Legora, Factory, Chroma, Granola, and many more. (5) Before starting Blacksmith @adijayaprakash, @aayush_shah15 and Aditya were engineers at Cockroach Labs, Faire, and Superblocks, working across core database, data distribution, and search infrastructure teams. They have assembled a rockstar infrastructure engineering team and have the ambition to build a world-class company. (6) This is only the beginning. Beyond its current products, Blacksmith has a super-ambitious roadmap across the code infrastructure stack (which will they launch soon). The amount of code being created is accelerating. The infrastructure stack needs to accelerate too. We are honored to work with this team. It’s going to be fun!

Excited to share vlt 1.0 along with our hosted registries & ecosystem mirrors now GA! A drop-in npm replacement, built so nothing runs on your machine just because you typed install. → faster delivery → malware blocking at the registry layer → graph-native querying

AI writes all your code. Does it still work? That's the gap: coding agents are scaling output, but the same with bugs and incidents. The biggest bottleneck is QA: how can you make sure the code your agents just wrote doesn’t break everything? The best QA are the ones who know everything about your product. Today, we give that knowledge to Momentic’s most capable agent. Every Linear ticket, every PRD in Notion, every PR - Momentic knows all. In just the past few weeks, our agents analyzed 70k+ test failures and created 600 tests. Achieving a 73% PR merge rate, and it’s growing fast. We’re grateful for the trust of our customers, including Notion, Xero, Webflow, Retool, Runway, Bilt, and more.

Basically, it’s cheaper than GitHub Actions — and it’s actually agentic. Loved the brilliant feature that automatically analyzes execution time for every workflow & job, then smartly adjusts the runner size. One of the coolest agentic experiences I’ve had with any product recently! @useblacksmith nice work!!

Blacksmith is part of @Redpoint's 2026 InfraRed 100 today. We started Blacksmith because CI needed its own cloud. Just over two years later, we're running millions of CI jobs every day for some of the fastest-growing companies. The same primitives we built to power CI will power much more than just CI. There’s a lot more coming :)

One of my deep-seated beliefs is that happiness is a choice. Obviously there are some natural temperaments and life circumstances that make being happy easier or harder, but I think you have to tell yourself your happiness is in your control. Even outside the big pillars of life like health, relationships, and work, we can have a lot of control on our daily internal experience with things like: - choosing to reframe losses as learning - being happy for others' success instead of jealous - looking for the good instead of bad in people - focusing on what we're grateful for vs. what we lack - looking forward to good things vs. dreading bad things - etc.

another underrated testbox use case: get agents to stress a flaky test across a dozen blacksmith VMs in parallel and let them hammer away until the cause is understood and deflaked. latest coding models are great at this kind of long-horizon work

Been so CPU-constrained on OpenClaw work. Switched local tests running to @useblacksmith and IT IS SO GOOD. codex can literally spin up to 32vCPU instances and rip through our test suite.

We've been getting asked about Mac runners a bunch. Today they're live on Blacksmith. Funny timing: one of our customers ran on Cirrus, which got acquired by OpenAI and shut down last week. We already had runners in production and got them migrated on the same day.