I've been working on something that started as personal frustration and turned into what I think might be a different way of thinking about AI agents.
The problem: Repetitive work is unavoidable if you want to grow at anything. Building an audience, managing research, shipping products, etc. they all require those soul-crushing repetitive tasks. The 1% inspiration, 99% perspiration thing is real.
I tried n8n, Make, Zapier. They work, don’t get me wrong, but the setup time doesn't make it worth the output once you're automating multiple workflows. They're deterministic but brittle and not adaptive. The tight coupling means one change breaks everything. The Agent-based ones were more like interns than employees: they won’t work if you don’t prompt (and in a particular way or style), couldn't generalize, and didn't fit how real teams actually work.
So I started asking; what if you could onboard an AI agent the way you onboard a human? What if it could pursue long-term goals autonomously, without you prompting it every step? What if it could be deterministic without being brittle? You know? Better output with a more seamless input.
That's why we built Compose AI.
Instead of you prompting the agent, the agent prompts you. Each agent has a job description (like you'd write for a human hire) that defines expectations and outcomes. You onboard them on your data - apps, drive, APIs, databases, documents - and they build a knowledge graph to understand relationships across your organization without explicit “tight-coupling".
Key differences:
- Autonomous goal pursuit: Tell an agent what to accomplish, not how to do it. It creates its own task list and executes.
- Human-like onboarding: Create agents using regular job descriptions containing your workflow and expectations.
- Multi-channel interaction: Email, voice, soon phone and Teams. You're not locked into a UI.
- Local browser automation: For tasks requiring a browser, agents can connect to your local computer browser. They work on a separate workspace on your computer - your browsing data stays local. Cloud browsing is also possible.
- Stack integration: Works across Google Docs, Notion, CRMs, whatever you're already using.
You don't need to be technical or be an expert at prompt engineering.
You just need a clear job description, which includes your workflow and expectations. The system handles the rest.
We have some pre-configured digital employees you can activate immediately (with job descriptions defined by us), or you can create your own. I see this as inverting the current paradigm - instead of humans adapting to how AI wants to be prompted, AI adapts to how humans already organize work.
Maybe it's a glimpse into how human-machine collaboration evolves. Maybe it's just a better way to stop doing repetitive tasks. Either way, I'd love for you to try it and tell me what breaks.
Hi HN,
I've been working on something that started as personal frustration and turned into what I think might be a different way of thinking about AI agents. The problem: Repetitive work is unavoidable if you want to grow at anything. Building an audience, managing research, shipping products, etc. they all require those soul-crushing repetitive tasks. The 1% inspiration, 99% perspiration thing is real.
I tried n8n, Make, Zapier. They work, don’t get me wrong, but the setup time doesn't make it worth the output once you're automating multiple workflows. They're deterministic but brittle and not adaptive. The tight coupling means one change breaks everything. The Agent-based ones were more like interns than employees: they won’t work if you don’t prompt (and in a particular way or style), couldn't generalize, and didn't fit how real teams actually work.
So I started asking; what if you could onboard an AI agent the way you onboard a human? What if it could pursue long-term goals autonomously, without you prompting it every step? What if it could be deterministic without being brittle? You know? Better output with a more seamless input. That's why we built Compose AI. Instead of you prompting the agent, the agent prompts you. Each agent has a job description (like you'd write for a human hire) that defines expectations and outcomes. You onboard them on your data - apps, drive, APIs, databases, documents - and they build a knowledge graph to understand relationships across your organization without explicit “tight-coupling".
Key differences: - Autonomous goal pursuit: Tell an agent what to accomplish, not how to do it. It creates its own task list and executes. - Human-like onboarding: Create agents using regular job descriptions containing your workflow and expectations. - Multi-channel interaction: Email, voice, soon phone and Teams. You're not locked into a UI. - Local browser automation: For tasks requiring a browser, agents can connect to your local computer browser. They work on a separate workspace on your computer - your browsing data stays local. Cloud browsing is also possible. - Stack integration: Works across Google Docs, Notion, CRMs, whatever you're already using. You don't need to be technical or be an expert at prompt engineering.
You just need a clear job description, which includes your workflow and expectations. The system handles the rest.
We have some pre-configured digital employees you can activate immediately (with job descriptions defined by us), or you can create your own. I see this as inverting the current paradigm - instead of humans adapting to how AI wants to be prompted, AI adapts to how humans already organize work. Maybe it's a glimpse into how human-machine collaboration evolves. Maybe it's just a better way to stop doing repetitive tasks. Either way, I'd love for you to try it and tell me what breaks.
Try it here: https://composeai.io
Looking forward to your feedback - especially the brutally honest kind.