The agentic AI sector is moving quickly, and a great deal is being written about it. Some of that analysis and research holds up under scrutiny; much of it doesn’t quite survive contact with the real world. One reason is that deploying agentic AI inside a business is far more complicated than many assume or wish to believe. Investors weighing where to back, vendors thinking about how to position, and corporate buyers evaluating what to procure all benefit from analysis grounded in how decisions actually get made. Information Matters exists to write the version that does.
We are an independent analyst firm covering the agentic AI sector from a business perspective rather than a technological one. We publish quarterly market reports, weekly news commentary with a thesis, and thematic deep-dives on the segments we believe matter most — coding agents, evaluation and observability tooling, voice infrastructure, vertical AI in legal, health and finance. Our research is built on a proprietary company database of more than 600 agentic AI vendors that we maintain in-house, refresh weekly, and use as the analytical substrate behind every piece we publish.
We work to a small set of disciplines. Where we estimate rather than measure, we say so explicitly. Where we take a position, we show the evidence. Where we are wrong, we change our view in writing.
Methodology
The methodology shapes what we publish. A few principles run through everything we do — most analyst firms hide their methodology behind a paywall; we don’t.
Primary sources only. We cite earnings calls, regulatory filings, company announcements, and first-party developer and CIO surveys. We don’t cite aggregator market-research firms as primary data, and we don’t pad our figures with CAGR extrapolations applied to opaque baselines.
The foundation-model attribution rate. When we size the agentic AI market, we apply a 25–35% attribution rate to foundation-model revenues — the share of API and product usage we believe is genuinely agentic rather than generative or chat. The range is built from four cross-checks: Anthropic’s product breakdown, the a16z CIO survey, OpenAI’s product mix, and Salesforce’s agentic-work-unit telemetry. It is the single most consequential assumption in our quarterly reports and the most likely to be challenged. We monitor the rate every quarter against the same primary sources and will revise the band — or the methodology behind it — when the evidence calls for it. Each report explains the rate in full and flags any change since the previous quarter.
Reconciliation, where we differ. When our market-size figures differ from comparable published ones, we explain the difference rather than ignoring it. Differences usually come from scope, layer or methodology — and the explanation is in every quarterly report’s reconciliation table.
A proprietary company database. Our research is grounded in an internal library of more than 600 agentic AI companies, classified by segment, region, stack role (agent, enabler, substrate, adjacent) and funding stage. The library is the analytical substrate behind every report and post we publish; companies covered in our reports get public profile pages on this site, while the wider library is internal to our research team.
What we don’t do. We don’t make investment recommendations on specific securities. We don’t publish vendor-sponsored editorial under our analyst byline — sponsored content is published separately and clearly flagged (see the Team page for how this works). We don’t apply Magic-Quadrant-style multi-axis vendor rankings that imply more precision than the data supports.
The hypotheses behind our work
Our analysis is shaped by a small number of working views about how agentic AI will play out across business and the wider economy. They are open to revision as the evidence develops — but they inform what we cover, the questions we ask, and the positions we take.
We expect the effective deployment of agentic AI to require a detailed understanding of business practices that vary across sectors and firms. The same underlying technology produces very different outcomes depending on the operating model it lands in — a coding-agent rollout in a software firm has different unit economics, governance patterns and adoption velocity from a customer-operations-agent rollout in a contact centre, even when both rely on similar foundation-model substrates.
We expect agentic AI to be transformational for most businesses, large and small — but the diffusion of the technology to take longer than many vendors and commentators are currently claiming. The structural bottleneck is enterprise pilot-to-production conversion, not the underlying capability. Headline run-rate revenue figures from foundation-model providers reflect API consumption; transformed enterprise workflows lag well behind.
We expect the business models of many enterprises to be radically modified within three to five years as the rules of competition change and value creation shifts. Companies that adapt early to a stack where agents handle coordination, decision-execution and customer interaction will compete on different terms from those that don’t. Incumbent valuations and challenger trajectories alike are still under-priced against this shift in many markets.
And we treat the question of AI sovereignty as central rather than peripheral. The agentic AI initiatives in China, Korea, India and the EU are creating parallel stacks with their own funding, regulatory frameworks and deployment cadences. Multinational enterprises will be navigating which stack to use in which jurisdiction within twenty-four months. This is not a footnote to a US-led market.
We hold these views openly and revise them as the evidence accumulates. Anyone disagreeing with any of them — or with how we apply them in a specific report — is welcome to write to us.
Who reads us
Our readers are senior — investors at venture funds and growth equity, vendor founders and product teams, CIOs and CTOs evaluating agentic AI for their organisations. The free side of the site covers our news commentary, thematic pieces, the Twelve to Watch rolling watchlist, and the company profiles for vendors we have covered. The Agentic AI Dispatches lands in subscribers’ inboxes every Friday. Paid access — the full quarterly reports, advanced library access, named-analyst inquiry time — opens later in 2026.
If you want to know what is actually happening in agentic AI, who is winning the segments worth caring about, and where the next year’s market is heading — we are written for you.
Subscribe to The Agentic AI Dispatches · Browse our reports · Connect with the team
