AI Impact Week Day 1: Why AI Adoption Isn’t AI Impact

Adoption is solved. Impact isn’t.

Here’s the thing. Almost every developer is using AI now. More than 90% of them, in fact. At the top companies, agents are writing half of all pull requests. So yes, adoption is done.

But ask an engineering leader what all that AI is actually worth, and you’ll probably hear something like this: “We rolled out the tools. Adoption looks good. I think it’s working. I can’t prove it.”

That’s the problem we spent AI Impact Week digging into. The way software gets built is changing fast. Work has moved out of the IDE and into terminals, agents and plan files. Some of your “contributors” aren’t even people anymore. And the metrics most teams rely on were built for a world that doesn’t exist now.

Meanwhile, your CFO, your board and your CEO have stopped asking “Are the developers using AI?” They’re asking “What are we getting for it?” If you can answer that, you’re in great shape for 2027. If you can’t, you’re spending money and hoping. And hope is not a strategy. 

Day 1 brought together Jellyfish leaders, two of the best-known voices in developer experience, our research team and a couple of customers to talk about exactly that. Here’s what you missed.

1. Intelligence across your AI-native SDLC

Andrew kicked things off with a bold claim: the code-writing part is basically figured out. Most code should be AI-written now. The real mess is the process around it, like who does what, what’s done by people versus agents, and how work gets handed off. That’s why AI feels like magic for one developer but often falls flat for a whole team. Jellyfish sees this across 1,300+ companies and 276,000 developers. More than half of those companies now have agents creating 22% or more of their PRs. And spending more isn’t the answer, because past a certain point extra tokens don’t buy extra output. Andrew boiled it down to three questions every team needs to answer: Where do I stand? Am I transforming? What is it worth? Then he added a fourth that nobody has cracked yet: how will all of this keep changing?

2. AI adoption isn’t impact: Rethinking engineering outcomes in 2027

Laura and Nathen’s take was simple: getting people to use AI was the easy part. Now the hard stuff is showing up. Teams are building things just because they can, not because they should. Developers are getting AI licenses along with sky-high expectations for output. And code review has turned into a traffic jam, because it was built for code written at human speed and it’s been asked to do way too many jobs. Their advice? Stop showing adoption numbers in exec meetings. Track how many experiments you run and how fast you learn from them. And never trust a single number on its own, because as Laura put it, it “will mislead you.”

3. Three AI impact questions: Where do I stand, am I transforming, what is it worth?

Krishna picked up Andrew’s three questions and showed how Jellyfish is answering them. Whatever AI tools you’ve bought, he said, you’ll get asked these questions, and in this order: Where do I stand? Am I transforming? What is it worth? So that’s how Jellyfish built its platform. It all starts with one view of your work where you can see people, people using AI and fully autonomous agents side by side. Agents count as contributors, so they show up in your throughput and cycle time like anyone else. The point, Krishna said, is that next time someone asks what you got for your AI spend, you don’t have to say “I think it’s working.” You can show them.

Where do I stand?

• Lifecycle Explorer: Shows you where time really goes from idea to production. That way you don’t make the most expensive mistake in AI right now: buying more code generation when code review is what’s slowing you down.
• AI Cohorts: Groups your developers by how they use tools like GitHub Copilot, Cursor and Claude Code, so you can see who’s getting value and who needs help.
• AI Adoption: Shows how your teams are using agents.
• Metrics Explorer: Splits your output between people and autonomous agents so you can see AI’s impact at a glance. You can also describe a custom metric in plain English, and an agent builds it for you.
• Research Insights: Lets you see how your AI use compares to 1,300+ other companies, right inside the platform.
• Blueprints: Ready-made templates for the AI metrics that matter, organized by the question you’re trying to answer.
• Jellyfish Assistant: Just ask. It can dig into the root cause of a problem or build you a quick dashboard.

Am I transforming?

• Skill Adoption: Shows which skills and practices are catching on, and which teams have them and which don’t.
• Behavioral metrics: A new kind of metric, like how often people let the agent use tools versus stepping in themselves. It tells you whether teams are working well with AI or just spinning their wheels. Two teams can spend the same and ship the same number of PRs, with one getting better every week and the other going in circles. Now you can tell which is which.

What is it worth?

• Understanding AI Cost: All your AI spend in one place, even though every provider reports it differently. It also helps you figure out whether you should be buying tokens or seats.
• AI spend-to-work attribution: Shows not just how much you’re spending, but what you’re spending it on, down to the projects and roadmap items.
• Pattern detection: Spots what your highest-performing teams do differently, automatically.
• AI cost benchmarks: Compares your spend to the market, not just to last quarter. That turns the conversation with finance from “this costs a lot” into “here’s what we’re investing in and why.”

Krishna also gave a sneak peek at what’s next: simple tools for individual developers. They’ll show you your own skill usage, how risky a change is before you make it, and how to write better prompts.

4. AI engineering trends: The research panel

Jellyfish has data on over 100 million PRs, 90 million prompts and 150 trillion tokens, and the panel used it to explain why using AI still isn’t the same as getting value from it. Teams are merging about twice as much code. But they’re only shipping 40–50% more features at best, and sometimes none at all. Speed up coding, and the slowdown just moves somewhere else: people’s attention, infrastructure, trust or process. The biggest gap is trust. At a typical company, agents create about 20% of PRs, but only about half a percent go all the way to production without a human. On cost, the panel’s advice was to stop thinking of AI as just another dev tool and start thinking of it as extra engineering capacity. Count people and AI together, and the cost of each PR is actually going down.

5. Customer spotlights: Restaurant365 and Daxko

Two engineering leaders talked about what this looks like in real life. Restaurant365 rolled out Claude in a big way in May and admits it “fell into the trap of adoption” at first. Now the team cares about “depth over frequency.” They look at things like PR size, cycle time and how developers actually use the tools, then check whether the spend lines up with features, quality and stability. When it didn’t, they switched to smaller, more focused bets. Daxko went AI-first across the whole company and reports AI usage, spend and business results to its board every month. Bill says the Data Hub and Jellyfish Assistant won over even the skeptics on his team, and AI Cohorts help him see who’s thriving and who needs a hand.

6. Closing remarks

Stop saying “I think it’s working.” Start transforming.

Your board and your CFO are already asking what you’re getting from AI. Jellyfish helps you answer with real numbers: where your teams stand today, whether things are actually getting better, and what your AI spend is worth, compared against 1,300+ other companies.

Request a demo and we’ll show you how it works.