How this index classifies AI adoption maturity, what evidence it runs on, and how it stays current.
The State of Play tracks how far AI practices have actually moved into real-world use, across 19 functional domains. Every practice is placed into one of six maturity tiers based on the evidence we can find for it, not on hype or vendor claims. The goal is a living, honestly-sourced answer to one question: is this AI capability something people are actually doing, or just something people are talking about? See How It Works for how a nightly scan actually finds and judges that evidence, and tonight's rotation.
Each tier answers a different question about a practice, in order from least to most mature:
Papers, benchmarks, and prototypes. Nobody has put it to work yet, not even in an early pilot at a named organization.
Experimental, but past pure research: early pilots, promising demos, and credible product launches, with production use rare or unproven. This is where the failures show up too: abandoned pilots, incidents, and hard lessons. When most of what gets reported is failure and disappointment, that's the blood in the water, and the practice stays here.
Real deployments at forward-leaning organizations, with named results. Early movers are getting value, but it's patchy: strong outcomes in some places, mixed ones in others, and most of the sector hasn't started. The tooling works but often takes real expertise, and the playbook isn't settled.
The ecosystem is mature enough that a competent team could adopt it today: GA tooling from major vendors, analyst recognition, and independent case studies with measurable results. The question is no longer whether it works but how to roll it out. Adoption is growing but isn't yet the majority.
Majority adoption, shown by survey data or case studies across different industries, not just the usual early movers. Not adopting it now needs a reason, and resistance is fading.
Assumed background — so widely adopted that nobody writes “should we adopt this?” articles about it anymore. It's stopped being a decision at all; attention has moved on to whatever comes next.
Promotion is qualitative, not a box-checking exercise: a practice only moves up when the evidence clearly answers the next tier's question above, with specific, named, independent evidence — not general impressions of momentum. The default is not to promote. When the case is unclear, the practice stays where it is until the next scan brings clearer evidence. Demotion is never automatic; it is a human editorial decision.
Every practice carries an evidence sidecar — a running list of the case studies, product launches, research papers, industry reports, and other citations that justify its current tier. Each item is scored for source credibility: engineering blogs with real deployment detail, peer-reviewed papers, and official documentation sit at the top; industry analyst reports, tech journalism, and vendor blogs in the middle; personal blogs, press releases, and social media at the bottom. A tier promotion needs multiple, independent, high-credibility sources — not volume.
A nightly research scan works through a 14-day rotation across the site's domains, so any given domain is re-examined every two weeks. Our newsletter follows the same rotation, so it covers a domain group shortly after that group's scan runs.
This index holds 339 practices and 774 recorded tier moves. The 737 moves dated before March 24, 2026 were assessed retrospectively in early 2026, working back through dated evidence; the 37 since then were recorded by the nightly scan as they happened.
Both kinds of tier move are judged against the same tier definitions above.
Alongside tier, each practice carries a trend: its momentum toward the next tier.
Evidence is building toward the next tier, and a promotion looks close. A promotion always starts a practice here.
Ordinary progress: nothing unusually strong or unusually thin for a practice at this tier. Most practices are here most of the time, so the index shows no badge for it. A few steady practices also carry a narrow-market note, where the use case itself is small and steady is the lasting answer.
The evidence shows momentum cooling: a vendor pulling back, a deployment paused or scaled down, adoption figures slipping. An early flag, not a verdict.
A specific obstacle has stopped progress toward the next tier, shown by more than one piece of evidence about this practice. The practice's detail view names the obstacle.
Real contraction: vendors exiting, adoption falling in successive reports, or a different approach taking over.
A blocked practice names what is in the way: regulation, technical ceiling, cost, trust and safety, or organizational readiness. Apart from a promotion, which sets Accelerating at once, a trend changes only when two separate scans agree, so one unusual night doesn't move it.
The State of Play's data and text are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). You may share and adapt them for any purpose, including commercially and for training AI models, as long as you credit The State of Play and link to thestateofplay.ai. The evidence we cite belongs to its publishers; follow each link for the original.
The State of Play is written and maintained by Beck Eco as part of a joint venture with Perly Consulting.