Dust.tt Enterprise vs Sup AI

“The buyer here is crystal clear: it's the IT or security team that's been blocking the AI project the line-of-business team has been begging for. SSO, RBAC, and audit logs aren't features — they're the unlock code for enterprise procurement. The wedge is smart: land with one Slack agent, expand into every department's knowledge base. The risk is that the 'contact sales' pricing wall means we have no idea if the unit economics survive a real enterprise deal with professional services and compliance reviews baked in. If they can hold a $30-50 per seat number without collapsing into custom contracts, this is a real business.”

“The direct competitors are Glean, Guru, and — increasingly — Microsoft Copilot Studio, which ships with the SSO and audit logs already baked into a tenant most enterprises already pay for. Dust wins if and only if the no-code agent builder is genuinely more capable than what IT admins can stand up in an afternoon with Copilot. The scenario where this breaks is a Fortune 500 with a Microsoft EA — the IT admin has Copilot Studio free in the bundle and zero incentive to add another vendor. What kills this in 12 months is not a competitor, it's platform consolidation: Microsoft and Salesforce both ship 80% of this natively and enterprises stop evaluating point solutions.”

“Extraordinary claims require extraordinary evidence. A 7.41 point jump on HLE via ensembling — without publishing methodology — smells like benchmark gaming. The latency of running 339 models in parallel is also a real concern for anything other than async research tasks.”

“The primitive is an agent-scoped RAG pipeline with an enterprise auth layer bolted on — that's a real thing, but the 'no-code' framing immediately raises the question of what happens when the agent needs to do something the drag-and-drop builder didn't anticipate. The DX bet is that IT admins, not engineers, are the deployers, which means the API surface for developers who want to compose this with their own tooling is probably an afterthought. There's no public API docs linked from the blog post, no mention of a SDK, and 'scoped to internal knowledge bases' tells me nothing about how document ingestion actually works at scale. I'll change my verdict the day there's a repo or a curl example in the docs.”

“The HLE claim needs independent verification, but the underlying ensemble approach is architecturally sound for factual Q&A tasks. Running 339 models is expensive — pricing will be the gating factor for production use. The $10 free credit is a fair trial.”

“The job-to-be-done is precise: let a non-technical team deploy an AI assistant over internal docs without giving up on compliance. That's one job, and the SSO plus audit log bundle is exactly what makes that job completable — without those two things, no enterprise IT team signs off. The onboarding question I can't answer from the announcement alone is whether a new user can go from SAML config to a deployed Slack agent in under 30 minutes, or whether there's a professional services call hiding in the middle. The specific product decision that earns a ship is scoping agents to internal knowledge bases by default — that's an opinionated choice that removes the biggest enterprise objection before the customer even raises it.”

“Model ensembling is an underexplored direction in the race to reduce hallucination. If Sup AI's approach scales, it could be more durable than fine-tuning individual models — you get the wisdom of the crowd across model families, training data, and architectures simultaneously.”

“For creative work, ensemble outputs tend to regress toward the mean — you get the most-agreed-upon version of something, which is usually the least interesting version. This is a tool for factual accuracy, not creativity. I'd stick with a single strong model for writing.”