AI now writes 41% of all code. Code churn is set to double in 2026. Delivery stability has dropped 7.2%. Microsoft is bundling the future of work for ~$70 a seat. Glean is at $200M ARR. And the tools engineering uses to measure itself were built for a world that no longer exists.
· · ·
TL;DR
- AI is writing 41% of code, doubling code churn, and dropping delivery stability — making engineering faster and less stable at the same time.
- The dashboard companies — Jellyfish, LinearB, Swarmia, Faros — can’t fix this. They’re read-only. They observe problems they can’t act on.
- Microsoft Frontier, Glean, and Google’s Gemini Enterprise Agent Platform are racing to build the AI operating system for knowledge work. None of them are building one for engineering specifically.
- Engineering needs its own operating layer: read every tool, surface cross-tool risk, run the response across the stack, learn from outcomes.
- PulseBoard AI is building it. Private beta open to ~10 engineering orgs at Series B–D scale through end of 2026. Request access →
· · ·
Two stats that should change how you think about engineering
In the time it takes to read this article, AI assistants will have generated about a million lines of production code somewhere in the world. According to recent industry data, AI tools now write 41% of all code and save developers 30–60% of time on routine tasks. Code churn is expected to double in 2026, and delivery stability has decreased 7.2% according to Google’s 2024 DORA report.
Read those two sentences again. AI is making engineering faster and less stable — at the same time. The throughput goes up. The quality of the work going through the system goes down. And nobody’s measurement framework is built to see the relationship.
AI is making engineering faster and less stable — at the same time. The throughput goes up. The quality goes down.
The DORA team noticed. In 2025, DORA officially replaced the elite/low performer tier framework with profile clusters and added a fifth metric. AI improved throughput 30–40% but increased delivery instability. The old “elite performer” framework is officially dead.
Read that one again too. The single most-quoted measurement framework in software engineering — the one your VP of Engineering put in last year’s board deck — was officially retired because the AI era broke it.
If you’re a CTO running a real engineering org right now, here’s the uncomfortable position you’re in: AI is changing the shape of your engineering work week by week, and the tools you use to understand your engineering org are reporting on a reality that ended 18 months ago.
This is the story everyone is missing while they argue about Copilot.
· · ·
Meanwhile, everyone is racing to build the AI Operating System
Look at what’s launching in the next sixty days.
Microsoft Frontier (E7), May 1, 2026. Microsoft 365 E7 brings together Microsoft 365 E5 for secure productivity, Microsoft Entra Suite for identity and access, Microsoft 365 Copilot for AI in the flow of work, and Microsoft Agent 365 as the control plane to govern and scale agents. The branding is on the nose: Frontier. Microsoft is positioning the entire bundle as the operating system for the post-AI enterprise. Every employee. Every tool. One control plane.
Glean, doubled to $200M ARR in nine months. Glean achieved a $7.2 billion valuation after doubling its ARR to $200 million in nine months, driven by an enterprise AI platform that has fundamentally moved beyond search. Their pitch: a permissions-aware knowledge graph as the foundation, agents as the layer above it. They’re not selling search anymore. They’re selling the cognition layer.
Google’s Gemini Enterprise Agent Platform. Google Cloud used Cloud Next 2026 to announce the Gemini Enterprise Agent Platform, an end-to-end workspace for building, governing, and scaling AI agents, while also framing the market as entering the “agentic era.”
OpenAI’s Workspace Agents. Same play, different vendor.
The shape of all of these is identical: read everything the company runs on, build a unified context layer, run agents on top, govern centrally. They are all reaching for the same prize — be the operating system for the modern company — and they’re attacking it from the knowledge-and-productivity side.
The market is ready for them. Futurum’s 1H 2026 Enterprise Software Decision Maker Survey found that 38.8% of enterprise buyers expect GenAI to be delivered primarily via agents, and 65.9% of enterprises follow a platform-first approach.
So here’s the question that should be keeping engineering leaders awake. If Microsoft, Glean, and Google are all building the operating layer for knowledge workers — who’s building it for engineering?
If Microsoft, Glean, and Google are all building the operating layer for knowledge workers — who’s building it for engineering?
Because nobody is. And the gap is about to matter more than at any point in the last decade.
· · ·
The dashboard companies aren’t the answer
You might point at Jellyfish, LinearB, Swarmia, Faros, DX. You’d be wrong, and the reason matters.
These companies built the engineering analytics layer over the last five years. They’re good at what they do. LinearB works with your Git, project management, incident, and release management tools to generate a DORA metrics dashboard quickly and easily for teams and the entire organization. Jellyfish reports engineering investment to your board. Swarmia surfaces flow metrics. Faros aggregates everything in one place.
But every single one of them is a read-only dashboard. They observe. They report. They surface trend lines. The action loop — now what do you do about it — happens entirely outside the tool, in a human’s head, in a meeting, in a Slack message someone has to type by hand.
This was fine when engineering was slow enough that humans could close the loop themselves. It is no longer fine. Watch the pattern repeating across every other category in 2026:
Traditional dashboards — static, retrospective, and manually configured — are being replaced by AI-driven operations analytics systems that do far more than visualize data. They predict outcomes. They recommend actions. Increasingly, they execute decisions automatically.
Rather than logging into dashboards, clicking buttons, and monitoring rule-based automations, enterprises are deploying autonomous systems that interpret goals, gather context across applications, make decisions, and act on them.
This isn’t a marketing trend. It’s a structural shift. Sales operations is moving from dashboards to agents. Customer support is moving from dashboards to agents. Finance is moving from dashboards to agents. Even operations analytics itself is moving from dashboards to agents.
Engineering is the only function still arguing about whether the dashboard is enough.
Engineering is the only function still arguing about whether the dashboard is enough.
· · ·
What an engineering operating layer actually looks like
Stop thinking about the engineering stack as a collection of tools and start thinking about it as a system that needs four operations running continuously across all of them.
Read. Connect every tool engineering actually uses — Linear or Jira, GitHub, Slack, Notion, Productboard, PagerDuty, the CI/CD layer. Build one normalized model of how delivery is flowing. This is the boring foundational work that everyone underestimates and that the dashboard companies got right.
Understand. Surface the signals that no individual tool can see. Decision latency. Dependency depth. Phantom progress. Blocker patterns. Coordination tax. The signals that actually predict whether a milestone will slip live in the gaps between tools, not inside any one of them. This is where the dashboards stop and where the cross-tool reasoning models can finally pick up.
Act. Execute the response across tools. Update the ticket. Post the structured update. Escalate the blocker. Draft the summary. Multi-step transactions handled atomically, with full audit, fully reversible. This is the part that no engineering analytics company has ever shipped, and it is the part that turns a smart dashboard into an operating layer.
Learn. Every cycle’s outcomes feed back into the recommendation engine. Retros stop being meeting transcripts and start being institutional memory. Pattern history sharpens every next prediction. The system earns trust sprint over sprint instead of demanding it on day one.
The closed loop is the thing. Every other category — sales ops, customer support, finance — is racing to close their loop with AI. Engineering still has open-loop tools and a worsening AI problem.
· · ·
Why this is a 2026 question, not a 2028 question
Three forces converge right now that didn’t converge in 2023.
Connector ecosystems are finally mature. Stable APIs across the engineering stack. OAuth flows that work. Middleware companies like Merge and Nango that make integration tractable. Reading every engineering tool reliably is no longer a five-year engineering problem. It’s a buyable building block.
LLMs can reason over messy cross-tool context. Pattern recognition across signals from many sources. Root-cause synthesis with confidence calibration. Production-grade reliability for the first time. The reasoning needed to close the loop didn’t exist three years ago. It does now.
Buyers are fatigued with point AI tools. VPs of Engineering have already bought sales AI, support AI, code AI. They are not asking for another disconnected agent. They are asking for connection. The thing that ties their entire engineering operation together. Tool fragmentation is now a board-level conversation.
If you wait until 2028, the window has closed and Microsoft has bundled everything generic. If you build for engineering today, you have a window — and it is short.
· · ·
The unsexy truth about AI in engineering
Here is what nobody pitching at AI conferences will tell you. The bottleneck in engineering productivity in 2026 is not the speed of writing code. AI fixed that. AI tools now write 41% of all code and save developers 30–60% of time on routine tasks. The code-writing problem is over.
The bottleneck is everything around the code. The blocker that nobody noticed for three days. The dependency that wasn’t communicated. The decision that took two weeks because the right person was on PTO. The ticket that got stuck in review. The launch that slipped because three teams thought another team was handling it.
The bottleneck is everything around the code.
These were the bottlenecks before AI. They are the bottlenecks now. AI made them more relevant, not less, because every hour saved on coding is an hour exposed to coordination overhead. Faster code production into a coordination-bound system produces more invisible queueing, not less.
This is the work the engineering operating layer does. Not “write code faster.” Not “summarize the standup.” The work nobody else is doing — make the system of engineering visible, predictable, and self-correcting in an era when the human-coordination layer can’t keep up with the AI-coding layer.
· · ·
What happens next
By the end of 2026, three things will be true.
Microsoft will have won the horizontal AI productivity layer for the average enterprise. Frontier and Agent 365 will be the default. Most companies will pay the per-seat tax and move on.
Glean and a small number of others will have won the knowledge layer. Knowledge graphs, permissions-aware retrieval, agentic search across documents and conversations. Their moat is real, and their renewals will hold.
And one company — maybe more — will have won the engineering operating layer. The connected loop across the engineering stack. The thing that reads every tool, surfaces the risk no single tool can see, and runs the response across the stack. The thing that makes engineering observable, predictable, and increasingly autonomous in an era when AI is breaking the old mental models of how delivery works.
We’re building it. PulseBoard AI is the operating layer for engineering — the connected loop across Linear, GitHub, Slack, Notion, Productboard, and the rest of the stack. We read every tool, surface the delivery risk that lives between them, and run the response across the stack — closing the loop from signal to action in one place.
The dashboard era is ending. The companies that ship a real engineering operating layer in the next twelve months will define the next ten years of how software gets built. The companies that ship dashboards will be quietly acquired or quietly forgotten.
You don’t need to take my word for any of this. Code churn is expected to double in 2026, and delivery stability has decreased 7.2%. The numbers are public. The trend is unambiguous. AI is breaking the engineering measurement framework faster than the engineering measurement framework can adapt.
The question is who builds what comes next.
The window is open in 2026. It will not be open in 2028.
· · ·
Join the PulseBoard AI private beta
If you run engineering at a Series B–D company and you’ve felt the pain we’ve described — the morning ritual of stitching status across five tools, the launch that slipped because the signal was sitting in a different system, the delivery risk that nobody saw forming — we are looking for design partners.
Private beta is open to ~10 engineering organizations through the end of 2026. Design partners get hands-on access to PulseBoard AI as we build, deeply discounted pricing, and a direct line to the founding team. In return, we ask for engagement and honest feedback.
If that fits, request access at pulseboard.ai/beta.
We’ll respond within 48 hours.
· · ·
Written by the team at PulseBoard AI. We’re building the operating layer for engineering — read every tool, surface the risk no single tool can see, run the response across the stack. If you’re working on this category from another angle, or running an engineering org and feeling the pain we described — we’d genuinely like to hear what you’re seeing on the ground. Get in touch.
The Dashboard Era Is Ending. Engineering Will Be the Last to Notice. was originally published in Level Up Coding on Medium, where people are continuing the conversation by highlighting and responding to this story.
