Replit: the factory infographic
Anyone can build an app by describing it; the agent writes, runs and ships the code.
What the company does
In normal terms
Agentic software creation platform that lets anyone build applications in natural language; millions of users and 500,000+ business users.
Replit's agent turns natural-language requests into working, deployed applications, so people who do not code can build software.
In lay terms
You tell Replit what app you want, like you'd tell a builder what house you want, and its AI builds it for you.
The risks it faces, especially from AI agents
In normal terms
- The agent touches production. A coding agent with deploy and database rights can change or delete live data while doing an unrelated task.
- Builders who can't review. Non-developers cannot judge what the agent changed, so mistakes and unsafe code ship unseen.
- Secrets and data in the app. Agent-built apps can embed keys, expose customer data or call services the owner never meant to use.
In lay terms
- The agent touches production. Asked to repaint a room, the builder knocks down a wall.
- Builders who can't review. If you can't read the plans, you can't spot the missing fire exit.
- Secrets and data in the app. The house gets built with the spare key taped to the front door.
How RiskMandate helps
- Reachwhat it can actually access or affectWhat the agent can change: code, databases, deploys, connected services.
- Mandatewhat the business authorisedWhat the builder actually asked for, in their words.
- Gapthe difference: the riskChanges beyond the request, such as deleting data or new services.
- Barrierswhat really stands in the wayCheckpoints, rollbacks and separate environments, named with evidence.
- Provenancewho owns it, who reviewed it, which versionWho asked, which agent version built it, who approved the deploy.
In normal terms
For Replit's business users, RiskMandate makes each agent's licence to operate explicit: what it may touch, what it may not, and what stops it. That is the record a company's security team or insurer asks for before letting staff build with agents.
In lay terms
RiskMandate is the building permit: it says what the builder may change, and checks nothing else was touched.
A first line for the conversation: Your users build with agents. Their security teams will ask what those agents may touch. We answer that, per agent.
Sources and method
The facts in grey (what it does) are from the company map, which cites:
- websummit.com, read 2026-11-01: LIS26 appearance page: name, country, industry, website
- websummit.com, read 2026-11-01: listed on the 2026 event partners list with its elevator pitch (the "what" field)
- websummit.com, read 2026-11-01: Michele Catasta, President, Replit: public speaker listed by Web Summit for LIS26
- aman-alliance.org, read 2026-11-01: Lusa, 25 Jun 2026: Michele Catasta, President and Head of AI at Replit, announced as speaker
The risks and how RiskMandate helps are our analysis, written by dev@web-summit.indi on 3 November 2026 from those facts and RiskMandate's model as this site explains it (Reach, Mandate, Gap; an Agent Behaviour Policy in practice). They are not the company's own statements and name no incident. A person reviews each before it is used.