The 2025 DORA report— Google’s annual research on how software gets built — landed with a single finding that reframes most of the AI conversation: AI is an amplifier. It does not turn a struggling team into a strong one; it makes strong teams faster and struggling teams more visibly stuck. After surveying nearly 5,000 technology professionals, the researchers concluded that the AI revolution in 2026 is less about the model and more about what the model lands on.
Lova is the chat-first project management tool where AI agents work as first-class teammates on a shared board. Read through the DORA lens, the board is the amplifier: it decides whether AI accelerates real work or accelerates the mess.
Key takeaways
- Google’s 2025 DORA report found that 90% of technology professionals are now using AI at work, a 14-point jump in a single year, with a median of two hours a day spent inside AI tools.
- More than 80% of respondents said AI improved their productivity, yet AI adoption still showed a negativerelationship with software delivery stability — the same paradox DORA flagged in 2024.
- Stack Overflow’s 2025 Developer Survey of more than 49,000 developers found that 84% now use or plan to use AI tools — while only 29% trust the accuracy of the output, the lowest figure the survey has ever recorded.
- The DORA team’s headline: AI does not fix a team; it amplifies what is already there.The lever that decides which direction it amplifies is the system around the AI — specs, platforms, feedback loops, and the project board.
- In 2026, the bottleneck has moved. The model is no longer the constraint. The shared surface where work is described, claimed, and finished is.
What did the 2025 DORA report actually find about AI?
DORA — the DevOps Research and Assessment program housed inside Google Cloud — has been the closest thing software delivery has to a longitudinal study. Its 2025 edition, the State of AI-assisted Software Development, draws on nearly 5,000 survey responses and more than 100 hours of qualitative interviews. The headline statistics are the part everyone shared first: 90% AI adoption, more than 80% reporting productivity gains, a median of two hours a day spent inside AI tools.
The more important finding sits underneath those numbers. AI adoption now correlates positivelywith software delivery throughput — teams shipping AI-augmented work are shipping more of it. But the same adoption correlates negatively with software delivery stability. Change volume goes up, and without robust automated testing, version control, and feedback loops, the system breaks more often. Speed without a control surface is not progress; it is the same team, accelerating into the same wall.
Nathen Harvey, who leads the DORA program at Google Cloud, framed the through-line in a post-publication interview: “In well-organized organizations with strong practices, AI amplifies that flow and accelerates value delivery. And in fragmented organizations with brittle processes, AI will expose those pain points and bottlenecks.” The report calls this the amplifier effect. It is the single most useful way to read every other AI statistic in 2026.
Why does AI amplify some teams and break others?
The amplifier framing answers a question that has haunted every CTO since 2024: why do two teams with identical access to the same AI tools end up at such different places? The answer is not the model, the prompts, or even the engineers. It is the system the AI is plugged into.
Consider Stack Overflow’s 2025 Developer Survey: 84% of more than 49,000 developers reported using or planning to use AI tools, up from 76% the year before. Yet trust in the accuracy of those tools fell to 29%, an all-time low, and 66% said dealing with AI output that was “almost right, but not quite” was their single biggest frustration. Adoption is nearly universal. Confidence is collapsing. That is not a model problem — the models are demonstrably better than they were eighteen months ago. It is a system problem. Without somewhere structured to land, AI output stays at “almost,” and the cost of finishing it lands on whoever receives it.
The pattern shows up at the institutional level too. MIT’s NANDA initiative found that 95% of corporate generative AI pilots delivered no measurable P&L impact within their first six months — an indictment not of the technology but of the organizational scaffolding around it. Gartner reached the same conclusion from a different angle, predicting that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Three different studies. One conclusion: the AI is fine. The container is not.
We argued the same point from the ROI angle in why most companies see no ROI from AI agents. DORA 2025 supplies the mechanism: AI does not generate value in isolation. It amplifies whatever flow already exists. If the flow is clean, AI multiplies it. If the flow is broken, AI multiplies the brokenness.
How does the project board become the amplifier?
Here is the claim worth sitting with: the amplifier rule applies to the board the same way it applies to the codebase. A board where every task carries a clear outcome, defined acceptance, and a small enough scope to be claimed and finished becomes rocket fuel for AI agents — they read the spec, do the work, ship the result. A board full of one-line tickets, stale columns, and titles like “fix the thing in the thing” becomes a different kind of rocket fuel: it produces more of the same noise, faster.
Call it the amplified-board principle. Whatever your board is right now is what AI will give you more of next quarter. If your board is a graveyard of half-described intentions, AI will fill it with half-described work. If the board is the source of truth — structured, scoped, claimable — AI becomes a teammate that runs against it. DORA found this for code. It is just as true for the layer above the code.
This is why spec-driven workflows have moved from blog post to category in the last twelve months. We wrote about that arc in spec-driven development. The board and the spec are the same artifact viewed from two angles. The board says who is making the outcome true. The spec says what the outcome is. Strong teams already wrote good specs and good tickets; AI made them indistinguishable from machines. Weak teams wrote vague ones; AI made them indistinguishable from noise. The DORA report supplied the vocabulary. The fix is to write the kind of board AI can amplify in the direction you actually want.
There is a quieter consequence too. The same study found that AI’s impact on stability was negative whenever change volume outran the control system. Translated to project management: when AI produces three times as much work, the board has to be three times as legible, or the team drowns. The polished output piles up. The reviewers triage instead of deciding. The pattern even has a name now — we covered it in workslop. DORA 2025 is the macro view of the same phenomenon. Workslop is what happens on the ticket. Instability is what happens to the system.
What does AI-amplified project management actually look like?
Strip the vendor language out and the shape is consistent across the teams that are getting DORA-style amplification right in 2026.
- Every task is a small executable spec.Title, outcome, acceptance, context — written tight enough that an agent or a human can act without a second conversation. This is what makes amplification productive instead of corrosive.
- The board is the single source of truth.No parallel Notion doc, no shadow Slack thread, no “the spec is in my head.” If it is not on the board, it does not exist — for humans or for agents.
- Status moves through the board, not through meetings. Claim, ship, mark done. The audit trail writes itself. The need to convene to find out what happened evaporates.
- Agents and humans are first-class participants.Both can claim, both can ship, both leave the same trail. The board does not care whether the work was done by a person or a process — only that the outcome matches.
- Feedback loops are fast and visible.Tests run, results post back to the task, drift surfaces immediately. DORA’s instability finding is what happens when this layer is missing. Build it before scaling agent volume.
None of this is exotic. It is what good engineering teams have always wanted. The difference in 2026 is that the cost of not having it has gone vertical. When one engineer can ship the output of a team, the gap between “the board is in order” and “the board is approximate” stops being an inconvenience and starts being the ceiling on the entire organization’s output.
How does Lova apply the amplifier rule?
Lova is built so the board itself is the spec. Each task is a structured artifact — outcome, acceptance, context attached — not a sticky note with hopeful text on it. Agents claim tasks through the same API humans use, ship against the same acceptance, leave the same auditable trail. There is no separate dashboard for the AI workforce, no translation layer between the planning surface and the execution surface, because the whole point of the amplifier rule is that the more daylight you put between intent and work, the more noise the system amplifies.
That is the only board design that survives a DORA-style read of the next two years. Boards designed for humans staring at swimlanes were tuned for an era when work moved at human speed. When the median developer is spending two hours a day inside AI and 90% of the profession has joined them, that era is over.
Frequently asked questions
What is the headline finding of the 2025 DORA report?
That AI is an amplifier. Strong engineering systems convert AI adoption into faster, more reliable delivery. Weak ones convert it into more output, lower stability, and exposed bottlenecks. The same tool, the same model, two opposite results depending on what it is plugged into.
Does AI adoption actually hurt software delivery stability?
In aggregate, yes — for now. DORA 2025 found a negative relationship between AI adoption and stability whenever change volume outran the team’s control systems (testing, version control, fast feedback). The fix is not less AI. The fix is investing in the layer that has to keep up with the AI — including the project board.
Why is developer trust in AI falling while usage rises?
Because adoption and confidence measure different things. Stack Overflow’s 2025 survey shows 84% of developers using or planning to use AI tools, and a record-low 29% trusting the accuracy of the output. People keep reaching for AI because it saves time on the easy 80% — and lose trust because the remaining 20% is “almost right, but not quite,” which is the most expensive kind of wrong.
How does the DORA amplifier finding apply to AI project management?
Directly. If the project board is clear, scoped, and structured, AI amplifies it into more shipped work. If the board is vague and approximate, AI amplifies it into faster approximation. The amplifier rule does not stop at the IDE; it travels up the stack to wherever decisions about what to build are encoded.
Where can I read the 2025 DORA report?
The full State of AI-assisted Software Development 2025 is available as a free PDF from Google. Pair it with the 2025 Stack Overflow Developer Survey for the developer-trust counterweight to the productivity story.