Every company’s claims in its own words. Bluestuff graded by the same test as everyone else.
Sometime this year “agentic” became the most overworked word in iGaming. Every vendor is using it, usually to mean different things — and buyers are left to tell the claims apart on demo polish, which is the one thing everyone does well.
This page is our attempt at a fair map. Every company on it is described in its own words, from its own public materials, with a source and a date. We’re on the map too, graded by the same four questions we’d ask anyone else. Where a competitor has the strongest proof in the industry, we say so.
It was first published alongside our CoinGeek piece of 13 August 2026, and it gets re-verified quarterly. If we’ve described your company wrongly, tell us — the correction route is at the bottom.
Scope. Do agents run every layer of the operation, or one function?
Handoff. Do agents pass work to each other, so one’s output is another’s input, or does each run alone?
Production. Does agent output ship to production against real systems — real APIs, real player data, real compliance workflows — or does it produce drafts and recommendations?
Governance. Who approves what ships? If the answer is vague, in a regulated industry, walk away.
No question asks how many agents a vendor has. Coverage, handoff, production and governance are harder to fake than a count.
On the test: Enormous scope and genuine production on mature, proven stacks — with the AI largely assistive today: a command centre over a conventional platform, the agentic part still emerging. One exception answers the production question completely, in one vertical.
Positions GiG Assistant as the “intelligent command centre” of its ecosystem, uniting platform, sportsbook, logic and data products under one AI-driven interface; quotes partner go-lives in as little as 12–14 weeks across 30+ regulated markets.
Embedding AI across a mature modular platform stack.
AI framed as internal acceleration and CRM tooling across an established platform; separately, its non-AI shipping velocity — prediction markets: 50+ signed projects within three months of launch — shows incumbent distribution strength.
Embedding AI into its established platform and brands.
The named concession. The industry’s strongest autonomous-operations proof, described in plain language rather than the word “agentic”: AI priced and traded 48% of network bets in 2025, 60%+ by Q1 2026, and the entire 2026 World Cup cycle AI-traded — 100m+ bets, no manual intervention. One function, at a depth nobody else currently matches; sells its AI pricing onto operators’ own platforms via Odds Feed+.
On the test: Real handoff and real production — inside one function. The honest description is agentic, vertically. The buyer’s question isn’t whether these tools work (several demonstrably do); it’s who coordinates them, and what happens in the seams between one vendor’s agent and another’s.
CRM agents where an operator describes an outcome and the system plans, segments, builds and executes the campaign.
XpertOS, positioned as an “agentic CRM operating system” for regulated engagement, with embedded agents and an approval layer.
Pushing decisioning agents into player journeys.
Launched July 2026 as an AI-native premium-player development platform, backed by Bettor Capital.
Context-aware sportsbook engagement, deployed with ComeOn Group on its proprietary sportsbook platform.
Sports-native agent infrastructure for sportsbooks and media.
On the test: The architecture answers handoff and production well in principle. What this group has to prove — Bluestuff included — is production evidence at scale, because architecture claims are the easiest kind to make.
Positions ELRA Core as an AI-native foundation under a family of products.
“Your casino, our AI-native stack”: own PAM, game engine, payments and back office with named operational agents on live platform data. Casino-only by design.
Full-stack AI-native platform, and the same agent workforce deployed as a layer on an operator’s existing stack. Full graded entry below ↓
Full graded entry below ↓On the test: The deployment model that addresses the actual installed base — operators who will never replatform. Most platforms fail at AI not because the models are weak but because the estates underneath are illegible; a layer that first makes the estate legible, then switches agents on, is the only version of “add AI” that survives contact with a ten-year-old stack.
Agents operating across an operator’s existing PAM, KYC, payments, bonusing and fraud tooling, with regulator-specific audit exports and policy enforcement on every action.
Full-stack AI-native platform, and the same agent workforce deployed as a layer on an operator’s existing stack. Full graded entry below ↓
Full graded entry below ↓On the test: Today a framework rather than an operation — and long-term the most interesting architectural bet on the map: if AI-legible platforms become standard, everyone’s agents get better, and differentiation shifts to who operates them well.
Open-source AI-native platform framework launched by Blurify in July 2026: build the platform so agents can understand and modify it.
Do agents run every layer of the operation, or one function?
Full stack for operators launching or consolidating, and the same agent workforce deployed as a layer on the platform an operator already runs — their PAM, wallet and licences untouched. Two modes, one platform.
Do agents pass work to each other?
Agents pass work down a real chain — product to design to build to QA to review — one agent’s output is the next agent’s input, with the artefact trail to show it.
Does agent output ship to production?
Output ships against real APIs, real player data, real compliance workflows, on a GLI-certified platform connected with seven gaming boards across Africa and payment networks across 12 countries. Certificates and approvals are the kind of thing serious counterparties should ask to see — we show them.
Company-reported — see the interrogation belowWho approves what ships?
Agents execute; humans approve every step that goes live through the release gate. Nothing ships without operator approval. In a regulated industry we think that’s a feature to lead with, not a caveat to bury.
Our headline numbers are company-reported, and you should treat them exactly as you’d treat anyone else’s. The interrogation they deserve — and the one to run on every vendor on this page: resolution rate, not deflection rate. Escalation rate. How often a human corrects the output. What share of agent-built work reaches production. Human hours per deployment. And how responsible-gambling obligations are handled inside every automated conversation.
Every description on this page comes from the named company’s own public self-description — its website or dated trade coverage — linked at each entry with the date we last checked it. We re-verify the full page quarterly; the “last reviewed” date at the top is real. Group assessments are our editorial view, stated in prose; we don’t score or rank named companies, and the only company graded in full here is our own.
Named on this page and think a description is wrong or stale? Send the source: [email protected]. Corrections verified against a public source are updated within 10 working days and logged in the changelog below.
Run the four questions on us.
Book a sessionAll trademarks and company names are the property of their respective owners and are used for identification only. This page is opinion and information, not procurement, legal or investment advice.