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 for finding, synthesising, verifying, and preserving organisational knowledge. Mostly leading-edge: literature review, competitive intelligence, and knowledge management tools are maturing quickly with five practices actively advancing. The main constraint is hallucination risk — fact-checking and source verification still require human oversight in high-stakes contexts.
The headline: AI research and knowledge tools are deployed everywhere and genuinely save time, but they still confidently make things up — and the cost of not checking has shifted from embarrassing to expensive, with courts now fining firms for unverified AI citations.
Almost every organization now has AI summarizing documents, searching internal files, and drafting research — it ships built into Microsoft, Google, and the major productivity tools you already pay for. The capability question is settled; these tools work and save real hours. The dividing line is no longer who has the technology but who has the discipline to use it safely. A small group has built the review processes, audit trails, and clean underlying data to deploy AI reliably; the majority are bolting it onto messy data and trusting outputs no one verified. The gap between those two groups is where the risk now sits — and over the past two years, raw AI model improvements have done almost nothing to close it. The fix is organizational, not technical.
Courts are now putting a price on unchecked AI. Judicial fines for AI-fabricated legal citations passed $145,000 in a single quarter, and one case sanctioned every lawyer on both sides for citing cases that did not exist. A global tally now tracks over 1,200 such court incidents across more than 30 countries. If your teams use AI for anything that goes on the record, a mandatory human-review step (a person checks each output before it ships) is now a liability control, not a nicety.
The fabricated-citation problem is accelerating, not fading. An audit of 2.5 million medical research papers found over 4,000 invented citations — a twelve-fold jump since 2023 — and publishers acted on fewer than 2 percent. The research and reference material your AI tools draw on is itself increasingly polluted, so "the AI cited a source" is no longer evidence the source is real.
Single-query AI search lost momentum. Tools like Perplexity and ChatGPT search are now standard, but their core accuracy problem looks structural rather than something the next model release will fix — accuracy can quietly collapse as the document pool grows. Treat these as fast first-draft research, never as the final word.
Regulators moved from watching to acting. European medicines regulators formally recognized continuous AI monitoring as standard practice, the UK opened a testing sandbox for AI in drug safety, and the EU's AI Act produced its first fines for opaque AI systems. Expect "show your audit trail" to become a procurement requirement.
Cheaper, leaner AI search architectures are arriving. New approaches match the accuracy of expensive setups at a fraction of the cost — one cut costs 700-fold, another 95 percent. Before signing or renewing a large RAG (retrieval-augmented generation — giving the AI your documents to reference) contract, ask vendors to justify their architecture on total cost, not just accuracy benchmarks.
Verification is becoming table-stakes. Specialist fact-checking tools demonstrably work where general-purpose AI does not; expect customers, auditors, and regulators to start asking how you verify AI output. Budget now for a verification layer rather than retrofitting one after an incident.
The 2027 cancellation cliff. Analysts forecast that 40 percent of ambitious AI agent projects will be scrapped by 2027 — almost always over cost and governance, not capability. Scope your AI initiatives to bounded, supervised use cases with clear ROI, and be skeptical of "fully autonomous" pitches.
Better models will not save you. Eighteen months of AI improvement produced essentially zero reliability gain for production research agents. The failure modes — confident fabrication, accuracy decay at scale — are baked into how the technology works, so the answer is process and oversight, not waiting for the next upgrade.
Your data is the bottleneck, not the AI. Roughly four in five enterprise deployments fail, and most failures trace to messy, stale, or poorly organized source data rather than the model. The unglamorous work of cleaning and governing your knowledge base is what actually determines success.
The tools are polluting their own well. About 40 percent of web content is now AI-generated, and synthetic material is creeping into the sources AI tools rely on — degrading quality in a loop no vendor can fix for you. The safe use cases are bounded ones over data you control and trust.
Go deeper: the full Research & Knowledge 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.