Why I think AI will save agile
For years, "agile" has been an interesting word in organisations. On one hand many people use it to signify the type of organisation (fluid, nimble, quick, connected to the customer) they want; on the other it became taboo due to the many failed agile transformations we've seen happen.
These days I am having a lot of conversations with executives about AI and in particular about AI agentic engineering; and my feeling is that now, something unexpected is bringing that agile thinking back. Not a renewed appreciation for the values and mindsets that agile tried to introduce in organisations, but more the fact that Agentic AI engineering, a technology with no particular loyalty to any management philosophy, is making the old, slow way of working impossible to sustain, and the mindset agile always asked for is coming back as a side effect.
Here's what agile actually was, what happened to it, and what agentic AI is now handing back to us, whether we asked for it or not.
What agile actually was
When agile was born in the early 2000's, it was born with the purpose of making work better for teams, organisations and their customers. It was a mindset before it was ever a methodology, built on behaviours like respect, truth, transparency, trust, and commitment.
That mindset spread because it worked. By the mid-2000s, organisations building digital products faster than anyone had before started asking how to do it with more teams, kicking off what became agile's second wave: agile at scale. And by the end of that decade, a third wave arrived: not agile as a delivery approach, but agility as an organisational trait, the ability to adapt, meet customers where they are, and keep improving rather than settling. That third wave, organisational agility, is still what most companies are chasing today, whatever they choose to call it or not.
What went wrong
Three things killed the essence of it.
We turned a mindset into a methodology. Humans don't love ambiguity, so we built rulebooks: Scrum, SAFe, certifications, job titles like "Scrum Master." The outcomes agile was meant to produce, faster learning, better customer outcomes, more engaged people, got replaced by the goal of following the steps correctly. The tail started wagging the dog.
From there, agile became an industry. Once you need certified people to run a methodology, you've built a machine that profits from the methodology existing, whether or not it's producing the outcomes it was meant to.
And then agile became a cost-cutting exercise. The organisational design model Spotify gave us was built to enable collaboration and speed through small, trusted, cross-functional teams. What most companies copied from it was the org chart, in service of headcount reduction, while dropping the trust and autonomy the model depended on.
What agentic AI is handing back
Agentic AI engineering isn't reviving agile because leaders finally went back and read the agile manifesto properly. It's reviving it because it's making the old alternative, slow handoffs, protected ideas, distant customers, too expensive and too slow to keep doing. Here's what's coming back, mindset by mindset.
Small, trusted teams over org-chart structures
When an agent can build, test and reshape something in the time it used to take to book a meeting, the old relay race stops making sense. A spec gets written, an engineer builds it, a designer reacts to it three weeks later: that sequence can't survive when the loop has shrunk from sprints to minutes.
So teams get smaller and closer, not because a consultant recommended it, but because size has become drag. Amazon rebuilt its agentic AI division around teams of six to eight where 30 to 40 people used to sit, and the teams that redesigned the actual work reported throughput gains many multiples higher than the teams that only swapped in new tools. Salesforce is testing one and three-person units in place of the standing scrum team and says, refreshingly, that it doesn't have clear answers yet. That's not a weakness. That's the mindset agile always asked for: hold the future as plural, learn as you go, don't wait for certainty before you move.
Experimentation as the default, not a value on a wall
Trying an idea used to cost a sprint. Now it costs an hour. When the cost of testing something drops that far, organisations stop protecting ideas from scrutiny and start throwing them at reality constantly, which is exactly the continuous improvement instinct that lean called kaizen decades before agile existed: identify, try, learn, try again, never treat a process as finished.
Closeness to the customer, restored by default
When "what if we tried this" becomes "here's a working version" in the same conversation, all the internal translation agile was meant to remove, the interpreting, the diluting, the "let me check with the team and get back to you," simply isn't there anymore. The space where customer needs used to get lost between departments is gone, not because anyone redesigned the org chart to close it, but because there's no longer time for it to open.
Business and tech working from the same reality
Product, design and engineering used to hand a plan down a line and hope it survived the trip. Now they're looking at the same working prototype at the same time, arguing about the same real thing instead of three different imagined versions of it. That's the cross-functional collaboration agile always asked for.
Verification and judgement, not typing, as the real craft.
As agents take over more of the writing of code itself, the valuable work shifts to designing the systems that check it: the standards, the gates, the guardrails that decide what's good enough to ship. This is the same shift lean pointed to when it separated the person doing the task from the person designing the flow the task sits inside. The craft moves up a level, from doing the work to designing the system the work runs through.
The part that still depends on us
There's a trap hiding in this story, and it's the same trap that swallowed agile the first time.
Lean has the same history. It came out of Toyota's shop floor as a philosophy: understand value from the customer's point of view, remove waste, improve flow. When American manufacturers went to Japan to learn from it, most came home having missed the point entirely, and applied it as a cost-cutting exercise instead. The tools survived. Kaizen, kanban, gemba walks. The philosophy mostly didn't.
The risk with agentic AI is exactly this. It's entirely possible to use these tools to simply do the old, siloed, slow, distant way of working, faster. Smaller teams that were only shrunk to save money and not redesigned around how the work actually flows report a wave of new incidents and rework instead of real gains, because you can't retrofit trust and autonomy onto a structure that was never built for it. AI amplifies whatever it lands on. Clear systems get faster. Messy ones get messier, faster.
So the mindset doesn't come back automatically just because the tools demand speed. The economics are forcing organisations toward the shape of agile: small, trusted, autonomous teams working close to the customer, experimenting constantly, judged on outcomes rather than output. But it's still a choice whether you build genuine trust and agency into that shape, or just make everyone move faster inside the same broken system.
The system of work is up for grabs right now. Nobody's written the new playbook, which means nobody can hand you the wrong one either. That's the work: designing, on purpose, around the people doing it. That way when AI forces the shape of agility back into your organisation, you build the substance too, not just the speed.
Do it properly, and you get the org agile always promised. Do it lazily, and you've built a faster version of what was already broken.
That's the choice in front of every leader right now, and it's the same one it always was.