State of AI Demos, Mid-2026 Update: What Changed Since May

In May we published the State of AI Demos 2026 and a sourced benchmark reference. Four months later the category looks different enough to warrant an update rather than a revision. This post covers what actually happened between May 17 and September 10, 2026, with a link to each vendor's own announcement, and what it means if you are evaluating demo software in the fourth quarter.

Two ground rules, consistent with our methodology: every figure below is attributed to the organisation that published it, and vendor performance claims are labelled as vendor claims.

The four months in one table

Not on that list, because it happened earlier but matters now: Navattic announced Navattic Agents on March 9, 2026, Saleo published its AI Demo Agent page on January 9, 2026, and Karumi launched on Hacker News in late 2025 as a Y Combinator Fall 2025 company describing itself as an agent that operates a real web app in a shared browser session.

Every major vendor now sells an agent

In May the phrase "AI demo agent" belonged to a handful of new entrants. By September it is on the product page of every established interactive demo vendor. Walnut's playbook defines the category as "an autonomous, conversational, intent-driven product experience that guides a buyer through a personalized product story without a human sales rep present." Navattic's co-founders put the bet more bluntly: Randy Frank predicts that "the future of product content will be completely bespoke. No two experiences should look the same."

The word is now shared. The products are not. Three architectures compete under the same label, and the difference decides what your prospect actually sees:

1. Live-product agents operate the deployed application in a real browser and talk the buyer through it. This is how RaykoLabs works, using Playwright on Browserbase for browser control and real-time voice models for the conversation. Karumi describes the same approach. Nothing is captured, so nothing goes stale when the interface changes.
2. Captured-demo agents guide buyers through interactive demos that were recorded first. Supademo's AI Demo Agent supports voice and text in 50 or more languages, runs 24/7, asks discovery questions, and surfaces the matching captured demo, video or deck. It is a usage-based add-on on Scale plans and above; every workspace starts with 1,500 free credits, which Supademo describes as roughly 75 demo calls. Navattic Agents and Walnut's agent platform are built on their capture-based products.
3. Data-injection live demos run the native product with curated demo data loaded into it, so complex enterprise software looks complete for every buyer. Saleo pioneered this for sales engineers, added a chat-based, multilingual AI Demo Agent in January, and is now sold inside Consensus.

We wrote a longer definition, with a comparison table against interactive demos, demo videos and chatbots, on the new What is an AI demo agent page. The short version for buyers: ask which of the three you are being shown, and ask to see it on your own product.

Consolidation: Consensus, Saleo and Peel

The Consensus announcement of June 9 is the largest structural change of the period. In the company's words, "Saleo brings flawless, data-rich live product demos to the Consensus platform. Built on proprietary technology that injects tailored demo data directly into the native product, Saleo enables revenue teams to run live demos that reflect the buyer's actual reality." The release names SAP, Salesforce, Seismic, Clari and 6sense as Saleo customers and Oracle, Atlassian, Salesforce and SAP as Consensus customers. Financial terms were not disclosed.

For buyers, three practical consequences follow. A Saleo evaluation is now a Consensus evaluation, with Consensus pricing and packaging. The enterprise, sales-led end of the market has consolidated into one platform that spans demo videos, live data-injected demos and an agent. And the self-serve, per-usage end of the market, where RaykoLabs, Supademo and Karumi sit, is now clearly a different purchase. Our Saleo comparison and Consensus comparison reflect the new ownership.

The buying-committee data that changes the demo brief

Storylane's August study is the most useful new dataset of the period because it measures who is absent. From 2,021 recorded conversations across 637 mid-market and enterprise deals, the average buying account had about 23 known contacts in the CRM but only 2.3 people on calls; roughly one in five known contacts ever joined a call. Champions had frequently evaluated competitors before the first conversation. Economic buyers shaped requirements and timelines without attending. Security, legal and finance first appeared at contract stage, and procurement calls averaged 1.89 attendees.

Read against the demo motion, the study says the call is the minority channel. Four out of five people who influence the deal will only ever meet your product asynchronously, on their own schedule, without a rep. That is the brief an AI demo agent has to satisfy: answer the champion's question at the moment it is asked, give the absent CFO something she can evaluate in ten minutes, and leave the sales team a record of who asked what. It is also why we price per completed demo rather than per seat: the value is in the conversations that happen when nobody is on a call.

Vendor claims worth reading with the label on

Two numbers will circulate from this period, and both deserve their provenance attached. Navattic's co-founders write that interactive demos appeared on 0.2 percent of websites in 2020 and on 18 percent of B2B websites by 2026, citing Navattic's own 2026 State of the Interactive Product Demo report. Walnut's playbook cites Walnut platform data for 34 percent faster sales cycles, 32 percent higher conversions and 26 percent higher engagement with personalised, intent-driven demos; the article does not reference an independent source for those figures.

Neither number is wrong for being vendor data, but neither is an industry benchmark, and both measure click-through interactive demos rather than agent-led demos. Storylane's September 9 statistics post makes the same point in its title, "Sourced, Not Recycled", and our benchmarks reference exists for the same reason. If you see any of these figures quoted without a source in a vendor deck, including ours, ask for the report.

What it means for a buyer evaluating in Q4 2026

• Decide the architecture before the vendor. Live product, captured demo, or injected data determines maintenance cost, accuracy and what prospects trust. Everything else on the feature list is secondary.
• Test the unknown question. Ask the agent something not in the docs. The right behaviour is to say so and offer a human. Walnut's own framing applies: "The real question is not 'AI or live rep?' The real question is 'which job belongs to which?'"
• Model pricing at your traffic. Per completed demo, per creator seat plus credits, and sales-led enterprise contracts produce very different bills at 200 demos a month. Our Supademo comparison lays the models side by side.
• Check what the sales team receives. Transcript, screens visited, objections and a score should arrive without anyone asking for a recap.
• Ask what happens at the next redesign. For captured-demo agents, the answer is re-capture. For live-product agents, nothing.
• Treat the acquisition as a packaging change. If Saleo was on your shortlist, request Consensus pricing and the roadmap for the combined platform before assuming continuity.

What it means for RaykoLabs

Our position did not change; the neighbourhood did. RaykoLabs remains the live-product, voice-first agent with published, per-demo pricing and a self-serve pilot. What changed is that the definition we used in May is now the industry's definition, and two of the largest vendors have adopted the language. We take that as validation of the category and as a signal that the differences between architectures will matter more, not less, as buyers get used to the word "agent" on every pricing page.

Three things we shipped in response this week: the AI demo agent definition page, a methodology and data notes page that states how every figure on this site is sourced, and side-by-side comparisons with Supademo and Saleo. Try the agent on our own product from the live demo, then judge the architecture question for yourself.

What we will publish next

A short explainer on what the Consensus and Saleo deal means for a mid-evaluation buyer, an honest comparison of Supademo alternatives, and a support-side roundup of customer support automation platforms. One post a week, each with sources, and none of them editing what we have already published.