Topics / Running AI in an organization

How do organizations measure AI's impact?

The metrics organizations use to judge whether AI is working.

36 passages from 19 organizations; 29 checked against the source.

What the record shows

Organizations measure engineering impact with different yardsticks

Meta reports output per engineer. Uber's CFO reports code output for engineers. Airbnb reports pull request throughput, and its CEO reports the number of features and improvements shipped. Spotify reports pull request frequency.

From Meta, Balaji Krishnamurthy (Uber), Christopher Sanson (Airbnb), Brian Chesky (Airbnb) and Spotify.

Some measure time saved or backlog cleared rather than volume

Airbnb's CEO said what matters is not how many tokens you use but product throughput, design quality and how much ships, and Airbnb says it cut time from concept to delivery on key initiatives by as much as 60%. A Shopify engineer said a dependency backlog fell about 70% in the first 11 days of an agent workflow. Bank of America says EricaAssist cuts average call times by nearly one minute per interaction.

From Brian Chesky (Airbnb), Airbnb, Erin Son (Shopify) and Bank of America.

Stripe compares the same sellers with and without its AI tool, and also cites an undefined figure

A Stripe engineering manager said that in weeks they use the internal tool Kai, account executives produce 2x the sales activity and close 39% more deals than the same sellers in weeks they do not. Separately, Stripe's president said seller productivity has increased by 20%, without defining the measure.

From Sharadh Krishnamurthy (Stripe) and Will Gaybrick (Stripe).

Some headline gains come without a stated definition

Bank of America says 19,000 developers have seen efficiency gains of over 20%, and Meta says output per engineer rose 30% since the start of 2025; neither defines the measure in these passages.

From Bank of America and Meta.

These summaries are this site's, not quotations. Each was checked by a second reader, who confirmed it says no more than the passages it cites. The organizations' own words follow.

When Account Executives use Kai, they produce 2x the sales activity, create 17% more opportunities, generate 26% more revenue opportunities, and close 39% more deals when compared to the same sellers in weeks they don't use it.

Sharadh KrishnamurthyEngineering manager, Agent Foundation team · Leader, Stripe

Post: Stripe developer blog, Stripe developer blog, July 30, 2026.

  • 2x output change, account executives' weeks using Kai vs the same sellers' weeks without it, as of 2026-07
  • 17% output change, account executives' weeks using Kai vs the same sellers' weeks without it, as of 2026-07
  • 26% output change, account executives' weeks using Kai vs the same sellers' weeks without it, as of 2026-07
  • 39% output change, account executives' weeks using Kai vs the same sellers' weeks without it, as of 2026-07

Checked against the source [ Grade B ] Said in the last year Passage page

One is you start measuring how many human touches you still do to have the agent do the right thing. That's supposed to go down.

Patrick DeboisProduct DevRel Lead · Practitioner, Tessl

Context. Debois names the first of two metrics he has come to believe in for judging how productive a team is with coding agents.

Talk: Coding Agents Don't Scale Themselves. Neither Do Your Teams. The Rise of Agent Enablement, AI Engineer World's Fair 2026 (YouTube), July 2, 2026, at 09:38.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

it's not about how many tokens you use, it's about the throughput of your product and the quality of your product designs and how much you're shipping.

Brian CheskyCo-founder and Chief Executive Officer · Executive, Airbnb

Earnings call: earnings call transcript via Platform Aeronaut (third-party transcript site), earnings call transcript via Platform Aeronaut (third-party transcript site), August 6, 2026.

Checked against the source [ Grade A ] Said in the last year Airbnb: $12.2 billion revenue, FY2025 Passage page

Since the beginning of 2025, we've seen a 30% increase in output per engineer, with the majority of that growth coming from the adoption of agentic coding, which saw a big jump in Q4.

MetaPublished by the organization itself (earnings call), no named author

Earnings call: earnings call transcript via Platform Aeronaut (third-party transcript site), earnings call transcript via Platform Aeronaut (third-party transcript site), January 28, 2026.

  • 30% output change, since the beginning of 2025

Checked against the source [ Grade A ] Said in the last year Meta: $201 billion revenue, FY2025 Passage page

We're seeing even stronger gains with power users of AI coding tools, whose output has increased 80% year over year.

MetaPublished by the organization itself (earnings call), no named author

Earnings call: earnings call transcript via Platform Aeronaut (third-party transcript site), earnings call transcript via Platform Aeronaut (third-party transcript site), January 28, 2026.

  • 80% output change, year over year

Checked against the source [ Grade A ] Said in the last year Meta: $201 billion revenue, FY2025 Passage page

Our PR throughput inside of Airbnb is 65% higher than it was before we introduced agentic AI.

Christopher SansonProduct Lead, AI Developer Experience · Leader, Airbnb

Social post: DX podcast page (DX sells developer-productivity measurement software) carrying a talk given at its DX Annual conference, DX podcast page (DX sells developer-productivity measurement software) carrying a talk given at its DX Annual conference, June 22, 2026.

  • 65% output change, agentic AI era vs before

Checked against the source [ Grade B ] Said in the last year Airbnb: $12.2 billion revenue, FY2025 Passage page

seller productivity has increased by 20%

Will GaybrickPresident, Technology and Business (also styled President of Product and Business) · Executive, Stripe

Interview: Tokens Are the New Dollars | Stripe's Will Gaybrick & David George (podcast episode: "Stripe’s AI Strategy: Build More, Not Less"), The a16z Show (podcast and YouTube channel of Andreessen Horowitz, an investor in Stripe), August 17, 2026. Reported by David George.

  • 20% output change, as said 2026-08-17

Checked against the source [ Grade C ] Said in the last year Passage page

And so I started tracking the number of iterations, the number of review rounds between when the pull request was opened and when it was merged. Sure enough, very little difference. … Once again, very little if any statistical difference between human-generated and AI-generated pull requests in terms of how many iterations before they're ready to merge.

Daksh GuptaCo-founder · Executive, Greptile

Talk: AI-Generated Code Is Already Competing With Human Code, AI Engineer (YouTube), September 27, 2026, at 07:57.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

You could figure out which ones took less tool calls, … which one made less errors, which one produced the correct outcome, and deterministically know what's going on. The point is you can start building data-driven systems without ever writing a single line of code. … in order to build a programming language, it wouldn't have taken eight people. It wouldn't have taken less than two years. It would have taken hundreds and thousands and tens of thousands of man hours, and then you would still have a broken system. And today, we can just spend billions of tokens and make it work, and we can make it stable.

Vaibhav GuptaCo-founder and Chief Executive · Executive, Boundary (BoundaryML)

Context. Gupta is comparing what building a programming language would have taken before with what it takes the team now.

Talk: Fighting Slop with Slop, AI Engineer (YouTube), July 31, 2026, at 6:00.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

Before I trusted this workflow, I back tested the scoring against months of clean up history. Then I ran it live. Seven for seven PRs with green CI. Now it runs daily on my laptop and hands work back for me to review. … $1.26 a pull request. A backlog of around 520 flags a year runs for under $700. If you had engineers working on this by hand, at least $26,000 that you're looking at.

Andrew OrobatorSenior Software Engineer · Practitioner, Reddit

Context. The speaker is a Reddit Android engineer. In the talk he describes this agent as running in his local workflow and on his laptop, and says some of his other practices are from his personal side projects. The $26,000 is his own estimate of manual cost, and the PR count is a small sample.

Talk: Scale the Judgment, Not the Model, AI Engineer (YouTube), June 30, 2026, at 11:59.

  • 7 flag cleanup prs green ci, the live run after back testing. A person approves
  • $1.26 flag cleanup cost per pr, per pull request. A person approves
  • $700 flag cleanup yearly cost, per year, backlog of around 520 flags. A person approves
  • $26,000 flag cleanup manual cost, per year, same backlog

Checked against the source [ Grade B ] Said in the last year Passage page

it grew like 7X over, the last, I think it was, two months, three months, something like that. And then you see our engineering headcount growth. It’s, gone up by, 10% or something.

Walden YanCo-founder and Chief Product Officer · Executive, Cognition

Context. Walden Yan was describing an internal graph of Cognition's merged pull requests, which he said had grown 7X; the passage is the continuation of that answer.

Podcast: The Age of Async Agents: Cognition's Walden Yan and OpenInspect's Cole Murray, Latent Space, May 28, 2026, at 00:03:01. Reported by Swyx.

  • 7 x merged prs growth, about two to three months, as recalled by the speaker
  • 10% engineering headcount growth, same period

Checked against the source [ Grade C ] Said in the last year Sells AI tools Passage page

And completely anecdotally based on my own experience, I've really only felt maybe 10 to 20% more productive with all of these phases that have come before. But now inside of Amazon, we've been running pilots with different teams across the company, and we've been seeing a median of 4.5x productivity improvement and sometimes more than 10x.

Clare LiguoriSenior Principal Engineer, AWS · Practitioner, Amazon

Context. Liguori is describing how pilots with Amazon teams compare with her own earlier experience of AI coding assistants.

Talk: From AI-Assisted to AI-Native: Building a Frontier Development Team, AI Engineer (YouTube), August 28, 2026, at 1:11.

  • 4.5x productivity gain multiple, Amazon pilots with teams across the company, as described in 2026

Not yet checked [ Grade B ] Said in the last year Amazon: $717 billion revenue, FY2025 Sells AI tools Passage page

So another experiment that I want to talk about is an experimental sprint that was done in the Prime Video Organization. They took a 10-day sprint and they … experiment where they put, again, six engineers in a room and they let them go wild with Kiro. They brought down the project delivery time estimate from what was going to be 90 weeks down to 24, based on all of the progress they had made in this 10-day sprint.

Clare LiguoriSenior Principal Engineer, AWS · Practitioner, Amazon

Context. Liguori is describing an experimental sprint in Amazon's Prime Video organization.

Talk: From AI-Assisted to AI-Native: Building a Frontier Development Team, AI Engineer (YouTube), August 28, 2026, at 4:05.

  • 90 weeks delivery estimate before, before the 10-day sprint
  • 24 weeks delivery estimate after, after the 10-day sprint

Not yet checked [ Grade B ] Said in the last year Amazon: $717 billion revenue, FY2025 Sells AI tools Passage page

They watched 50 teams that were totally … distribution of early career folks, mid-career senior engineers, and that worked on existing systems. … And they watched them for the better part of last year, and they found something super interesting. They found that there was a big difference in the productivity gains that they saw between half of the teams and the other half. And in this case, they used a productivity metric of deployment velocity to production. … And they saw that for half of the teams, they achieved less than 3X increase.

Clare LiguoriSenior Principal Engineer, AWS · Practitioner, Amazon

Context. Liguori is describing a pilot run by Amazon Stores, which she says covers Amazon.com, its retail websites and its physical stores.

Talk: From AI-Assisted to AI-Native: Building a Frontier Development Team, AI Engineer (YouTube), August 28, 2026, at 5:59.

  • 50 teams teams in pilot, watched for the better part of last year
  • 3x deployment velocity gain half of teams, Amazon Stores pilot

Not yet checked [ Grade B ] Said in the last year Amazon: $717 billion revenue, FY2025 Sells AI tools Passage page

[uReview] does … around 25,000 comments a week, and we get 10% of them actually get some feedback … We also saw that the overall address rate was around 67%, and almost three quarters of the high severity issues were usually addressed by the developers

Ameya KetkarSpeaker, AI Engineer World's Fair 2026 (job title not stated) · Practitioner, Uber

Context. Ameya Ketkar is reporting results for uReview, Uber's code-review system. The address rate is how often a developer acts on a comment, as he defined it earlier in the talk.

Talk: Building uReview, Uber's Multi-Agent Code Review Engine (talk, AI Engineer World's Fair 2026), AI Engineer (YouTube), August 28, 2026, at 10:13.

  • 25,000 comments ai review comments, per week. AI assists a person
  • 67% review comment address rate, as of the talk. AI assists a person

Not yet checked [ Grade B ] Said in the last year Uber: $52 billion revenue, FY2025 Passage page

we saw that against like a very naive implementation, our costs were down by 60%, and our quality and our accuracy was up by around 70%.

Ameya KetkarSpeaker, AI Engineer World's Fair 2026 (job title not stated) · Practitioner, Uber

Context. Ameya Ketkar is summarizing the effect of the observability and evaluation work on uReview, Uber's code-review system.

Talk: Building uReview, Uber's Multi-Agent Code Review Engine (talk, AI Engineer World's Fair 2026), AI Engineer (YouTube), August 28, 2026, at 10:50.

  • -60% ai review cost change, against a naive first implementation
  • 70% ai review quality change, against a naive first implementation

Not yet checked [ Grade B ] Said in the last year Uber: $52 billion revenue, FY2025 Passage page

requests and tokens tell us like usage, like merge requests and pipelines tell us about platform activity, but they do not tell us what value … providing. … the more meaningful long-term question really is like what is the future velocity and outcomes that we can measure with less friction, getting changes out safely.

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader, Cloudflare

Context. Asked how Cloudflare thinks about measuring return on AI spend.

Talk: From Assisted to Delegated: Cloudflare's AI Engineering Stack (interview, Background Agents Summit 2026), Ona (YouTube), May 6, 2026, at 20:49.

Not yet checked [ Grade B ] Said in the last year Cloudflare: $2.2 billion revenue, FY2025 Sells AI tools Passage page

The impact on developer velocity internally is clear: we’ve never seen a quarter-to-quarter increase in merge requests to this degree. As AI tooling adoption has grown the 4-week rolling average has climbed from ~5,600/week to over 8,700.

CloudflarePublished by the organization itself (post), no named author

Post: The AI engineering stack we built internally, on the platform we ship, Cloudflare blog, April 20, 2026.

  • 8,700 prs merged, 4-week rolling average per week, after AI tooling adoption grew (from about 5,600)

Not yet checked [ Grade B ] Said in the last year Cloudflare: $2.2 billion revenue, FY2025 Sells AI tools Passage page

we've increased the number of features and improvements we shipped this year by nearly 80%.

Brian CheskyCo-founder and Chief Executive Officer · Executive, Airbnb

Earnings call: earnings call transcript via Platform Aeronaut (third-party transcript site), earnings call transcript via Platform Aeronaut (third-party transcript site), August 6, 2026.

  • 80% output change, H1 2026 vs H1 2025

Checked against the source [ Grade A ] Said in the last year Airbnb: $12.2 billion revenue, FY2025 Passage page

19,000 software developers have experienced efficiency gains of over 20% through real-time coding assistance

Bank of AmericaPublished by the organization itself (post), no named author

Post: Bank of America newsroom (the company's own press site), Bank of America newsroom (the company's own press site), October 14, 2026.

  • 19,000 people engineers, as of 2026-10-14
  • 20% output change, as of 2026-10-14

Checked against the source [ Grade B ] Said in the last year Bank of America: $113 billion revenue, FY2025 Passage page

We have like 50 users and I don't want to scare people, but the PR throughput for these folks is like 3X higher again.

Christopher SansonProduct Lead, AI Developer Experience · Leader, Airbnb

Social post: DX podcast page (DX sells developer-productivity measurement software) carrying a talk given at its DX Annual conference, DX podcast page (DX sells developer-productivity measurement software) carrying a talk given at its DX Annual conference, June 22, 2026.

  • 3x output change, as of 2026-06-22

Checked against the source [ Grade B ] Said in the last year Airbnb: $12.2 billion revenue, FY2025 Passage page

we're not saying, "Wow, we need fewer sellers." We're saying, "So, we need a lot more sellers."

Will GaybrickPresident, Technology and Business (also styled President of Product and Business) · Executive, Stripe

Interview: Tokens Are the New Dollars | Stripe's Will Gaybrick & David George (podcast episode: "Stripe’s AI Strategy: Build More, Not Less"), The a16z Show (podcast and YouTube channel of Andreessen Horowitz, an investor in Stripe), August 17, 2026. Reported by David George.

Checked against the source [ Grade C ] Said in the last year Passage page

JSON: /api/v1/topics/measuring-impact.json