Uber burned its AI coding budget in four months. Most teams see that and focus on token costs. But a 50-person eng team can lose ~$700K/year to the coordination overhead that happens after the code is written. Read the latest Confident Commit: https://circle.ci/4rjo5SG

22s vs 69s. Same repo. Same agent. Same lint+test gates. We ran a controlled A/B inside chunk-cli comparing sidecar validation to push-per-task CI. The result: 3.1x faster feedback. LLM cost? Basically flat. Tokens tracked fixes, not wait time. Full breakdown ⬇️

AI made building cheap, but it didn't make owning cheap. Our CTO's three questions for build vs. buy in the AI era: -Does it touch what customers pay you to deliver? -What's the cost of being wrong? -What's the long-term cost of ownership?

Cloud changed infrastructure from fixed to dynamic capacity. Chris from @augmentcode.com thinks agents are about to do the same thing for engineering teams. Full convo: https://circle.ci/4uuNWc8

27 seconds vs 5 minutes. That’s what inner loop validation looks like with Chunk sidecars. Fast feedback in the inner loop that validates changes while your agent is still working. Cleaner commits in the outer loop. Available for all CircleCI users today! Learn more: https://circle.ci/43h8dWR

Engineering leaders are budgeting AI wrong. Running your team at 2x for 3 weeks, then throttling them back to 1x because they hit a token cap, isn't cost optimization. It's just introducing friction into the system. Full ep with Chris from @augmentcode.com: https://circle.ci/4uuNWc8

Toronto, we'll see you next week at CTO Craft Con! 👋 Find our booth to talk about how to keep pipelines moving at AI speed without constant human effort to hold it together & catch Rob discussing how to lead through market disruption. Join us: https://circle.ci/4u9pCMQ

Turns out Chunk exists outside the terminal too. You can now 3D print your own Chunk and keep it close for moral support when CI has other plans. Download files on Thingiverse: https://circle.ci/4cTPNkt

Introducing Chunk sidecars: fast, pre-configured environments that bring validation into the inner loop in 60 seconds or less. Microbuilds run the checks, providing faster, cheaper feedback in the inner loop & more green pipelines in the outer. Read on: https://circle.ci/4ncdrw4

📣 We’re hiring an AI Community Engineer! 📣 We’re looking for someone relentlessly curious, building agents just to see what happens, breaking things on purpose, and sharing what they learn.

Debugging failing CI pipelines is slow because bugs are hard to find. 🔎🐛 Chunk finds the issue, writes the fix, and opens a PR you can actually merge. Full tutorial: https://circle.ci/48ZdOEc

Most codebases are getting worse over time and AI is increasing the pace of change—making it easier than ever to see what’s working and what isn’t. 😬 Compound engineering is the alternative: systems improving with every change, so the work gets easier and the codebase gets stronger as you go.

GitHub, GitLab, or Bitbucket? 🧐 Turns out the answer says a lot about how your team ships. We pulled the pipeline data to find out: https://circle.ci/4rjo5SG

30 years of engineering experience, and Rob still thinks the most important thing a leader can do right now is use the tools. Full convo: https://circle.ci/47Z5Bj2

Your AI coding tools are shipping features faster than your test suite can keep up with. Luckily, Chunk can scan your codebase, find the gaps, write the tests, and open a PR. Full walkthrough: https://youtu.be/nbUHX3SVt9U

Chunk regrets to inform you those retries were not, in fact, different this time.

The fastest way to find your delivery bottlenecks isn't an audit, it's a team running at full AI speed until something breaks. Rob joined @thoughtworks.com to discuss, plus what our 2026 State of Software Delivery data shows about the teams pulling ahead ⬇️

Teams aren’t struggling to generate code, they’re struggling to make it all work together. We’re joining O'Reilly Media on April 16 to dig into how AI is actually shaping software delivery end-to-end:

AI delivery gut check: % to prod in 24h? Review time? Failure rate? That’s the bottleneck. See how you compare: https://circle.ci/3NASptw

Most engineering roadmaps are built around certainty, but that model is beginning to break down. When the way people interact with software is changing this quickly, exploration stops being a side project and starts becoming the day-to-day. Tune in: https://circle.ci/4sAZadU

More code doesn't always mean better outcomes. The teams winning have built systems AI can rely on, not just better prompts. Luca Rossi breaks down what AI maturity actually looks like. 👇

Not on Chunk's watch

Long test suites slow everything down. But by running only the tests that matter, we cut one project’s suite from 30 minutes to about a minute and a half, which makes feedback way faster! Learn how faster feedback loops drive more stable delivery: https://circle.ci/3YiQgEx

Your best engineers want to solve hard problems, not spend their day reviewing code they didn't write. The top teams in our 2026 research figured out how to make that possible: https://circle.ci/4aygjgQ

Your CI pipeline is the new bottleneck, but don’t fret! Chunk can fix it. Our autonomous CI/CD agent reads your build history → rewrites your config → opens a PR. You just merge it. Here’s the full demo: https://circle.ci/4dhM3df

Our State of Software Delivery Report sponsor, @thoughtworks.com, shared their take on the report: AI increases code throughput, but delivery systems haven’t caught up. More code ≠ more shipped software. Read their perspective: https://circle.ci/4sm03Y0

Chunk headed to Toronto for AgentCon this weekend and had a great time! 🇨🇦 We loved chatting with so many developers and catching a session from our team on agentic delivery and what happens when CI stays in the loop while AI agents build. Where should Chunk go next? ✈️

90% of our engineering team uses Claude Code daily. But the more meaningful number: a team of 8 built Chunk, our fully autonomous agent in just a few weeks. Closed loop, auto-triggered, validated PRs. @anthropic.com published the full story:

CLI vs. MCP for AI-assisted dev isn't an either/or. CLI = inner loop. Fast iteration, zero overhead, model already knows the tools. MCP = outer loop. Auth, structured JSON, cross-system coordination. Full breakdown 👇

Junior devs are generating code at senior speed. Senior devs are buried in PRs. The bottleneck didn't disappear, it just moved to your review queue. Read the 2026 State of Software Delivery Report: https://circle.ci/4aygjgQ

AI ships working software, but it doesn't always ship software people love. That gap is the moat now. More from Loïc and Rob here: https://circle.ci/4sAZadU

The goal isn’t just fixing incidents faster, it’s learning from them. While AI can surface insights, improvement only happens when you carry that learning into future code. Here's the full conversation: https://circle.ci/491BGr6

Want a peek at what we're shipping next? 👀 Join us at our live product roadmap session on March 5 to discover what we're shipping now, what's coming next, and how engineers are using CircleCI to move quickly (without breaking things). Register: https://circle.ci/4u7a8cr

This is what closing the AI delivery gap looks like. Chunk detects flaky tests, fixes them, validates the change, and opens a PR automatically. No manual cleanup required! ▶️ Demo: https://circle.ci/4oEksER

📣 PSA: Shipping faster is not the same as delivering value. If your sales, marketing, or legal teams aren't ready, AI just shifts the bottleneck instead of removing it. Watch the full convo to learn how teams are improving flow end to end: https://circle.ci/491BGr6

AI writes code at machine speed. Your validation doesn't. That gap is the AI Delivery Bottleneck, and it may be costing you! Our 2026 State of Software Delivery Report fully breaks it down: https://circle.ci/4aygjgQ

AI boosted feature branch activity across the board last year. 📈 But for the typical team, main branch throughput actually went down 7%. 📉 We dug into 28M workflows to understand why in the latest Confident Commit: https://circle.ci/49HdjQ7

Rob sat down with Luca Rossi to chat about one big takeaway from this year’s State of Software Delivery report: AI is increasing change volume, but only teams with strong delivery systems are converting that into shipped software.

The 2026 State of Software Delivery is here, and this year’s data makes one thing clear: AI has accelerated how fast we write code, but most delivery systems have not kept pace. Read the report: https://circle.ci/4aygjgQ

Even when AI speeds up coding, delivery becomes the limiting factor. Small improvements in delivery time can reduce queueing and keep work moving. Watch the full AI Is Breaking the SDLC session to see it in action: https://circle.ci/3YiQgEx

The biggest costs in CI rarely show up on a budget line. They show up as lost time, delayed delivery, and senior engineers being pulled into work that could have been handled earlier. When you model that impact, the economics change quickly. Find the full example: https://circle.ci/3L643vd

Most teams don’t notice this day to day, as it usually shows up as waiting on feedback, context switching, and stalled momentum. But over time, it can turn into a real cost for engineering teams. Full breakdown: https://circle.ci/4cdyXwT

When a test passes and fails on the same SHA, that’s genuine flakiness. Here’s how teams can surface it quickly and cut down recovery time. Full demo: https://circle.ci/4oEksER

This isn’t about working faster. It’s about autonomous validation removing the friction that pulls engineers away from meaningful work. When delivery systems adapt to each change, that time adds up quickly. Learn more: https://circle.ci/4ag5H5U

When you look at CI throughput growth by team performance, a clear gap shows up. 🏆 Top 5% of teams increased throughput by 97% 🚀 Top 10% grew by 47% ⚖️ Median team grew 4% AI is increasing development activity, but the gains aren’t evenly spread. We dig into why here: https://circle.ci/49HdjQ7

Ever wonder where the time goes in CI? ⏳ Rob walks us through how small delays in feedback, failures, and reviews add up to real cost for engineering teams.

Average daily CI throughput is up 59% YOY across all projects, the largest jump we’ve ever seen. 😮 What’s less obvious is how uneven those gains are and why more activity doesn’t always translate into feeling faster or more effective. Read on: https://circle.ci/49HdjQ7

Teams expected AI-assisted coding to lighten the load, but what if it's actually making software delivery more unstable? ▶️ Watch the full session of AI is Breaking the SDLC here: https://circle.ci/3YiQgEx

Just because an AI approach works at Google doesn’t mean it’ll work everywhere else. The best place to start is talking to your team about what’s working (and what's not). Tune in for a look at what it’s like to build at AI speed: https://www.youtube.com/watch?v=nXx52z2OukY

Ship Smarter Session: Connecting Cursor to CircleCI with MCP Learn how Cursor connects to CircleCI using MCP, end to end. Full tutorial: https://circle.ci/4b0Fhr0