By ByteBot
AI coding agents made writing code almost frictionless. You prompt, they push. Except now the bottleneck is no longer the code — it is the CI queue sitting between your agent output and your main branch. GitHub commits are on pace to hit 14 billion this year, up from 1 billion in 2025. Agent-generated pull requests grew 325% in six months. CI infrastructure built for human-speed workflows is not keeping up.
That problem just attracted 45 million dollars. Blacksmith, a San Francisco startup that builds purpose-built CI infrastructure for GitHub Actions, closed a Series B led by Peak XV Partners at a 550 million dollar valuation. That valuation was 60 million dollars less than a year ago. The 10x jump is not a coincidence — it is a timestamp on exactly when the validation gap became the dominant infrastructure problem of the AI coding era.
What Blacksmith Is Actually Selling
The product sounds deceptively simple: faster GitHub Actions runners on dedicated hardware. Single-line migration. Bare-metal gaming CPUs, Firecracker microVMs that boot in under three seconds, and a co-located cache running at 400 MB/s — four times faster than GitHub default backend. Docker builds run up to 40 times faster on cached layers. The company claims 2x the speed of GitHub hosted runners at 60% of the cost.
But the more interesting move is Codesmith, their new cloud coding agent. It does not just run tests faster — it diagnoses CI failures and auto-fixes them, keeping pull requests green without developer intervention. The next release, Codesmith QA, will autonomously test changes before they merge. Blacksmith is positioning itself not as CI infrastructure but as the entire validation layer between AI generation and production. Read the full announcement on the Blacksmith blog.
The Numbers Behind the Pain
The funding round is easier to understand once you look at the actual developer data. A 2026 Faros AI study tracking 22,000 developers found code review time up 441.5%, while task throughput rose only 33.7%. LinearB analysis of 8.1 million pull requests across 4,800 organizations found that developers feel 20% faster with AI tools but are actually 19% slower — a 39-point gap between perception and reality. AI-assisted pull requests wait 5.3 times longer for reviewer pickup than non-AI PRs.
The CI layer is where this shows up most clearly. During heavy AI agent load periods, GitHub Actions wait times peaked at 54 minutes. Normal is 15 to 40 seconds. That is not a scaling problem you solve by throwing more of the same hardware at it. SiliconAngle has additional context on the market dynamics.
Blacksmith own growth reflects the urgency. The company had roughly 700 customers at its Series A in September 2025. It has more than 6,000 now — an 8.5x increase in under twelve months. Customers include Supabase, Clerk, Ashby, Mercury, and Expensify. Some of its largest accounts spend over 1 million dollars a year. ARR went from 10 million dollars to tens of millions with a team that grew from 10 to 30 people.
What This Means for Your Pipeline
The 45 million dollars is largely going toward compute expansion. Blacksmith currently manages hundreds of thousands of cores and is targeting a 10x increase in capacity. That scale suggests they expect demand to continue compounding, not level off.
If your team has adopted any AI coding agent in the past year, there is a reasonable chance CI is now your pacing constraint. The metrics to check: average CI queue time, PR cycle time from open to merge, and how often PRs sit waiting for the first reviewer. If those numbers have gotten worse while your team feels more productive, TechCrunch analysis confirms you are in the majority.
The options are not limited to Blacksmith. Namespace leads on raw CPU benchmarks for GitHub Actions runners. Depot is better for Docker-heavy pipelines. Ubicloud is the cheapest managed option. But what Blacksmith is building toward — the combination of fast hardware and an agent that fixes what the hardware surfaces — is a more complete answer to the problem than faster runners alone.
The Ouroboros Problem
There is something worth noting in the structure of what Blacksmith is building. Their Codesmith agent generates code. That code needs to be tested. It gets tested on Blacksmith CI platform. AI development is now creating its own validation market, and that market is capitalized at over half a billion dollars nine months into its existence.
CEO Aditya Jayaprakash put it plainly: Validating code is still a bottleneck, and it is an even bigger bottleneck because people are writing even more. That is going to be true for a while. If your pipeline has not felt it yet, it will — the only question is whether your CI is ready when it does.
