The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI applied from user research through to shipped product experience. Wide maturity spread: A/B testing and analytics are established, prototyping and design systems are good practice, but nearly half the domain is bleeding-edge — generative UI, autonomous UX research, and AI-native product frameworks are experimental. Most practices are stalled, with more energy in tooling announcements than production adoption.
The headline: The tools that design and build your product now hand work to each other automatically, end to end. Three-quarters of enterprises have already had to pull an AI agent back out of production.
Most product and design teams now have "agentic AI" — software that acts on its own without being prompted — somewhere in their workflow. Over the past fortnight the major vendors wired those tools together: research findings flow into specs, specs into designs, designs into live code, with no person in between. A small group with clean data and real review discipline is getting genuine value. The majority are not: 88% of AI agent pilots never reach production, and most that do get rolled back. The reason has changed, and that is the important part. The problem is no longer that AI makes things up; it is that company data is fragmented and systems don't connect — 80% of rollbacks trace to plumbing, not to the AI. If you can't measure a payoff yet, you're in the pack. The teams pulling ahead are spending on data and controls, not on more tools.
The toolchain wired itself together. Figma shipped the ability to push design changes straight into a company's code repository without an engineering ticket; Anthropic launched Claude Design with a direct line into its coding tool; Builder released live multiplayer editing where designers, product managers and QA all edit production code at once. Ask which of your vendors now pass data automatically to AI agents, and who approved it.
The rollback paradox got sharper. Sinch surveyed 2,527 senior leaders across ten countries: 74% have already withdrawn a customer-facing AI agent after launch, yet 62% still have agents live and 98% are increasing AI spend. Companies with mature governance roll back more often, at 81%, because proper testing catches failures weaker organizations never see. If your AI programs have never had a rollback, that is more likely a measurement gap than a quality record.
The failure mode moved. An analysis of over 10,000 enterprise AI failures found hallucination — when an AI tool confidently makes things up — now causes under 10% of incidents, down from being the dominant concern two years ago. Breakdowns between systems account for 31%. The next dollar buys more reliability in data cleanup and integration than in a better model.
Session recording became a courtroom risk. A US federal appeals court revived a wiretapping claim against Bloomingdale's over software capturing customer keystrokes and page visits. California now has 3,968 such cases, 81% of the national total, and an audit found session recording running on 59 major US healthcare sites where patient data is in play. This software sits quietly on most corporate websites — ask legal what it records and what consent covers it.
Accessibility automation shipped as the underlying code got worse. Siteimprove, Evinced and Digital.ai all released tools that automatically find and fix accessibility problems. In the same window, practitioner testing found every sample of AI-generated interface code failed accessibility standards, because the AI reproduces the inaccessible code it learned from. Automated fixing is a speed multiplier on a problem your own AI tools are now creating faster.
A cancellation wave is scheduled. Gartner expects more than 40% of agentic AI projects to be cancelled by 2027 for want of demonstrable value, and only 18% of leaders are confident in controlling what their AI agents can access — while 40% already run them in production. Set the success metric and the access rules before the next purchase, not after the pilot.
Accessibility deadlines are firm and the legal defense has narrowed. US federal deadlines run to April 2027–2028, extended precisely because AI cannot yet automate the fixes, and European enforcement is already issuing fines. Of 3,948 US accessibility lawsuits filed in 2025, 983 hit sites that had already bought a compliance widget. Budget engineer time to fix source code and keep an auditable record — that record is what wins dismissals.
Renewals will expose shelf-ware. Around 40% of competitive-intelligence platforms are abandoned within twelve months on alert fatigue, and $50,000 experimentation contracts have been documented running two inconclusive tests in six months while comparable engines now sell for about $150 a month. Before the next renewal, measure usage per seat, not licenses bought.
Automating a handoff doesn't remove the work the handoff was doing. The checkpoints being deleted are where humans caught problems. Not one of 200 surveyed site-reliability leaders reports high confidence in AI-generated code after deployment, two-thirds of product managers still substantially rewrite AI-drafted requirements, and a study of 22,000 developers found output up 34% while incidents per code change rose 243%.
Better governance looks worse before it looks better. The organizations with the strongest controls report the highest rollback rates. Expect your own numbers to get uglier as measurement improves, and be ready to explain to a board why that means the controls are working.
The people who must enforce the rules are the ones who least want to. Most brand-voice programs fail not because tooling is missing but because writers won't adopt the checking layer into daily work. Design-system AI stalls near 95% compliance for the same reason: review capacity cannot keep pace with how fast the machines generate.
Go deeper: the full Product & Design briefing — the longer analytical write-up, plus every practice we track in this domain with its maturity rating, the tools to consider, and the evidence behind our assessment.