Dust

Meet Sarah B., our newest Forward Deployed Engineer in NYC! 🗽 We're celebrating Sarah's arrival by showing how quickly she's already making Dust work for her (and our customers). What's different about a Forward Deployed Engineer? Sarah writes code that ships to production, but also responds to customer support tickets, and gets hands-on with real Dust use cases. All in her first few weeks. Here's how she uses Dust in her own day-to-day: "I joined the engineering team three weeks ago, and our code review agent has been invaluable during onboarding. It's helped me align my PRs with our style guide and ensure my code fits our architecture - all without monopolizing senior engineers' time. Our deep-dive agent has been equally transformative. My first engineering tasks would have taken at least twice as long without it. I've been able to independently investigate implementation decisions and customer use cases, building context that would have required lengthy pairing sessions. What used to take days now takes hours. What excites me most is showing engineering teams how AI can actually improve code quality instead of degrading it. In an era where teams are rightfully concerned about AI-generated code, Dust proves that the right tools can raise the bar rather than lower it." We're building Dust to be the operating system for work in AI-driven companies. Excited to have Sarah on the journey with us!

would claude code work if you had to map out every step in a flowchart? obviously not. coding agents work because they reason in real-time, maintain context, and adapt to what you're doing, because they have agency. so why are we building business agents with flowcharts? at Dust, we wrote about what changes when agents can actually reason 👇 (blog post in the comments)

You love using AI agents but you hate spending time tweaking their prompts? At Dust we're building agents that can analyze their own feedback and suggest instruction improvements. Guided by data, approved by you. I'm excited to share this blog post on a topic I'm currently working on: Self Improving Agents 👇 https://lnkd.in/eQX8Wymn

ai can ship, your local setup can’t if 2026 engineering already feels nothing like 2024, that’s because it isn’t: agents parallelize the work, and infrastructure becomes the limiter we fixed that with dust-hive: isolated dev envs in seconds, built for parallel agents. blog post in the comments !

Your agents aren't just answering questions anymore. 👀 They're booking meetings, updating CRMs, calling APIs - acting like privileged employees at machine speed. Traditional security controls weren't built for this. Join Dust's security team for a 30-minute deep dive on what changes when AI moves from passive chatbots to autonomous action. Built for CISOs, Security Directors, IT Leaders and anyone responsible for securing AI in their organization. Register here: https://lnkd.in/eYpqKH7b

📢 Introducing Mentions.  AI agents that ping you when they need you. Here's what happens now: • Your teammates can @mention you directly in Dust • Your AI agents can @mention you when they need your expertise • Everything lands in one inbox so you can track what needs attention Sales agents pull you in when deals need a human touch. Support agents route questions to the right specialist. Automated reports notify stakeholders when results are ready. Work flows: human → agent → human, keeping momentum without losing context. Available now for all Dust workspaces!

Dust is helping Vanta reclaim 400+hours/week across their GTM organization through a connected network of Dust agents. When Daniel Baralt and Shashank Khanna came to us, their biggest blocker was that company expertise lived in silos. Compliance knew frameworks, finance had usage data, product owned customer feedback, but connecting those dots meant hours of manual work. They tested 7 platforms before choosing Dust. They liked that Dust struck the right balance -> simple enough for anyone to build with in 30 minutes, but powerful enough to scale programmatically across the organization. Other platforms were either too shallow for enterprise needs or too technical for widespread adoption. They built their system in 3 layers: 1/ Domain agents built by experts Each function created its own agent. GRC built a compliance expert. Finance created one for usage insights. Product developed a Voice of Customer agent. 2/ Cross-team orchestration QBR prep now automatically pulls from finance, GRC, and Voice of Customer agents to generate complete decks in minutes. 3/ Agents embedded in daily work The same agents operate in Slack. The GRC agent handles security questions directly in channel with quick human review, so specialists never answer the same question twice. As a result, they're saving thousands of hours annually while improving the quality of customer interactions and internal reviews. They've also discovered just how much appetite there is at Vanta for integrating AI into daily work. When they hosted their last company-wide Dust training, 180 people showed up. What began as a GTM initiative has now become a company-wide capability and we couldn’t be happier to be working with such a world-class team. 🫡

Doctolib's VP of Data & AI sunsetted their own internal AI project, DoctoGPT. It had 800 active users in a matter of weeks. Then the feature requests started flooding in… "We need connectors for Zendesk." "Can you add this integration?" Nacim Rahal, who built DoctoGPT, realized his healthcare engineers were becoming infrastructure engineers. "We created a Feature Requests JIRA board that revealed massive demand. We were overwhelmed with requests within days of launch." Every smart engineer debugging connection timeouts or new integration requests was an engineer not revolutionizing patient care. So Nacim became the first person to advocate for shutting down his own creation. His framework was elegant: "Build what's in our core business, buy what will be a side project." 1/ Maintaining AI infrastructure required resources across multiple teams 2/ Healthcare innovation was getting starved of talent 3/ They'd never match the pace of specialized AI platforms "We'd rather put 100% of our core resources on helping patients and solving practitioners' problems." Six months after switching to Dust, they're at 3,000+ users across the organization, with no engineering resources spent on AI infrastructure. All talent focused on healthcare features. "I was the first one to ask for it. We wanted to be free from the 'burden' of having to be the product owner, rather than the customer." Sometimes the bravest thing a leader can do is kill their own success. :)

Wakam, a leading European insurer, hit 70% employee adoption and shipped 136 AI agents on Dust. Legal cut contract analysis time by 50%. Here’s the exact playbook you can copy (we won't tell!) 1/ Executive sponsorship and business framing Executive leadership made AI agents a strategic priority. Adoption was framed around business outcomes, not experimentation, and reinforced in weekly company meetings. 2/ Platform first, model agnostic Wakam chose a platform that could evolve with the models and integrate cleanly with existing systems, including Notion, SharePoint, Slack, Snowflake, and HubSpot, with SSO and APIs in place. 3/ Dual-layer permissions by design Access is controlled at two levels: agents can only retrieve data from approved spaces, and employees can only use agents they are authorized to access. This enabled broad rollout without compromising sensitive data. 4/ Company-wide launch Agents were positioned as the preferred way to find information and complete work, not a side tool for early adopters or technical teams. 5/ Structured enablement at scale Wakam invested heavily in enablement before and after launch: training tailored to roles, weekly open office hours, a dedicated Slack support channel, clear internal documentation, a meta-agent to help employees build their own agents. 6/ Build at the edges Of 136 deployed agents, roughly 70% were created by business teams. The AI Engineering team focused on the most complex use cases and supported others as needed. 7/ Measure and reinforce momentum Internal dashboards tracked adoption by team, usage patterns, and productivity impact. Early wins, including a 50% reduction in legal contract analysis time, were shared internally to sustain momentum. 8/ Sequence capability responsibly Wakam started with knowledge assistants. Action agents followed once trust, permissions, and governance were in place. More autonomous agents are introduced only after legal, risk, and security review. Check out the comments for the full write-up. Thank you Wakam team for trusting us with your agents roll-out! Etienne DEBOST Sofía Calcagno

“I can't imagine working without this tomorrow” is what Spendesk’s customer support team said after trying out Dust. Leadership took the hint and rolled it out company-wide. 6 months later: 90% adoption. As Europe's leading spend management platform, they’re also a Payment Institution, which means letting employees use random AI tools is a no-no. They picked Dust, ran a POC with their CS team, and when the positive feedback was so strong, leadership expanded it to the entire company. Spendesk launched hackathons where teams competed to build the best agents and created an AI Champions program with 1-2 people per department getting 10% dedicated time to implement AI into workflows. “When you have a tool that can serve as a sandbox to build what you might otherwise buy, and it can benefit the entire company while meeting security requirements, the decision becomes obvious,” says their CEO Axel Demazy “It costs more not to adopt AI than to adopt it.” Thank you Greyg Sinigaglia de Malibran Michyl Culos Cécile H. Mary Barthe for sharing Spendesk's inspiring journey with Dust 🙌