Inspiration

Knowledge workers lose hours to repetitive web tasks - over 40% of workers spend at least a quarter of their week on manual chores like data entry and data collection. Non-technical users can’t easily automate browser workflows. Current solutions require coding or complex setup, and rigid scripts often break with any website change.

Project managers and researchers resort to tedious copy-paste between web apps (updating trackers, gathering info) because there’s no simple way to connect these tools. This wastes time and invites errors. Even today’s AI assistants can summarise or advise, but you still do the clicking and form-filling. There’s a gap in translating intent into action on the web.

What it does

Agenti lets you describe any web task in plain English - “check my email every morning and schedule calendar invites,” for example - and our assistant executes it on your behalf, just like a human personal assistant would.

How we built it

We shipped a server-rendered SaaS platform using Next.js (TypeScript, React) and integrated with OpenAI’s APIs. Our chat interface and backend orchestration live in one codebase, so we iterate fast.

Challenges we ran into

Maintaining conversational context across multiple steps, handling API rate limits and errors, and designing a UI that feels both powerful and simple took more effort than we first thought.

Accomplishments that we're proud of

In just a few weeks, we launched a working MVP, onboarded an initial waitlist, and demonstrated core workflows like email checks and calendar scheduling - all without writing a single browser extension.

What we learned

We discovered the nuances of conversational UX, the importance of robust error handling in multi-step automations, and the need for secure token management when connecting user accounts to external services.

What's next for Agenti

We’ll add deep integrations (CRMs, project tools), build a browser extension for on-page actions, and launch an agent marketplace so anyone can publish and share custom automations.

Built With

  • nextjs
  • prisma
Share this project:

Updates