A Quieter Rival to Microsoft’s AI Dominance
Microsoft Copilot arrived with enormous momentum – enterprise contracts, deep Office integration, and the full weight of the Microsoft ecosystem behind it. For most corporate IT departments, adopting it felt less like a choice and more like the default path. But a growing number of teams are quietly routing around it, choosing a Paris-based startup called Dust that takes a fundamentally different approach to how AI gets deployed inside a company.
Dust is not trying to be a general-purpose chatbot bolted onto existing software. Instead, it positions itself as a platform for building custom AI agents – purpose-built assistants that connect to a company’s actual data sources, internal tools, and workflows. The pitch is specificity over breadth, and for certain teams, that distinction is proving persuasive enough to replace something they already have.
What Dust Actually Does
The core product is a workspace where operations, support, engineering, and HR teams can each build agents trained on their own knowledge bases – Notion docs, Slack threads, Salesforce records, GitHub repositories, and more. Rather than one AI assistant that knows a little about everything, Dust lets a company deploy multiple agents that each know a lot about one domain. A customer support team’s agent can be trained entirely on product documentation and past ticket resolutions. A sales team’s agent can be built around CRM data and call transcripts. Each one operates within a defined context, which tends to make outputs more accurate and less generic.
That architecture is the direct counterargument to Copilot’s approach. Microsoft’s tool is deeply integrated with Word, Excel, and Teams, but it draws on broad language model capabilities rather than tightly scoped company-specific knowledge. For companies that live primarily inside the Microsoft stack, that integration is valuable. For companies running a more fragmented tool set – which describes most startups and many mid-size enterprises – Copilot’s tight coupling to Office becomes less useful, and Dust’s connector-based model starts to look more practical.
Dust currently supports model flexibility, allowing teams to switch between providers like Anthropic, OpenAI, and Mistral rather than being locked into a single underlying model. This matters more than it might initially appear. As model capabilities shift rapidly, the ability to swap the engine without rebuilding the entire workflow is a real operational advantage, particularly for teams that have learned to be skeptical of vendor lock-in after one too many platform migrations.
Why Teams Are Switching Mid-Contract
The churn from Copilot to Dust is not happening at the executive level, at least not initially. It tends to start with individual teams – usually product or engineering – that tried Copilot after the company signed an enterprise agreement and found that the tool did not fit how they actually worked. The mismatch is often less about raw capability and more about customization. Copilot offers configuration options, but those options operate within Microsoft’s structure. Dust is built to be configured from scratch, which appeals to teams that want to define what their AI agent knows, how it responds, and where it pulls information.
There is also a practical consideration around cost transparency. Microsoft’s Copilot licensing is bundled with Microsoft 365 at a per-seat price that covers the full suite. Companies that do not use most of the Microsoft stack are effectively paying for integration they do not need. Dust charges for the platform directly, which means teams can evaluate it on its own terms. That pricing clarity is part of what makes the comparison straightforward to make internally when a team wants to push back against a top-down software decision.
The Enterprise AI Market Is Not Winner-Take-All
The assumption embedded in Microsoft’s rollout of Copilot was that enterprise AI would follow the same consolidation pattern as productivity software – one dominant suite capturing the majority of the market by being good enough and deeply integrated. That logic held for email, calendaring, and document editing. It is proving harder to apply to AI assistants, because the value of an AI tool is much more dependent on what data it can access and how precisely it is configured for a specific workflow.
This is the same dynamic playing out across developer tooling – specialized platforms built for specific workflows are chipping away at general-purpose incumbents by offering tighter integration with how practitioners actually work. The pattern is consistent: a big platform owns the broad market, and a focused challenger captures the teams that need something more precise.
Dust is not the only startup occupying this space. There are competing platforms building similar agent frameworks, and several enterprise software vendors are adding their own AI layers. But Dust’s early traction – it has reportedly been adopted by a range of European and US tech companies – suggests there is real demand for a middle layer between raw AI APIs and fully locked-in enterprise suites. The companies that want to build sophisticated AI workflows without becoming dependent on a single model provider or software vendor are finding that Dust fits that gap.
The harder question for Dust is what happens when Microsoft responds with more configurability, or when one of the larger cloud providers decides to build a direct competitor with their own distribution muscle. The startup’s current advantage is focus – it does one thing and has built the product around that one thing with care. Maintaining that focus while adding the enterprise features that larger customers will eventually demand is the kind of balance that has undone more than a few promising B2B platforms before Dust had a chance to grow into them. For now, the teams defecting from Copilot are voting with their workflows, and that vote is loud enough to warrant attention.
