# Proactive Agents

> The definitive guide to proactive AI agents — what they are, how they differ from reactive agents, and how to build them.

This site is a working manual on proactive agents: AI agents that act without being prompted. A proactive agent wakes itself up when time passes, data changes, or a message arrives — rather than waiting for a human to invoke it. The site covers architecture, implementation patterns, and lessons from production.

## Key Concepts

- A proactive agent is defined by how it wakes up, not what model it runs or what framework it uses.
- There are exactly three triggers: time (schedules/intervals), change (data mutations via webhooks), and message (human or system communication).
- Proactive agents require three primitives: a wake-up mechanism, persistent state, and durability (checkpointing, idempotency, spend control).
- Reactive agents poll on intervals and forget between runs. Proactive agents receive push events and maintain state.
- The shift from reactive to proactive is the most important architectural decision when shipping an AI agent.

## Content

### Guide
- [What Are Proactive Agents? The Definitive Guide](/guide/): Comprehensive definition, architecture, comparison with reactive agents, and how to build them.

### Essays
- [Proactive agents need three primitives](/posts/three-primitives/): Why agents need a clock, a watcher, and an inbox wired together with durable state.
- [Reactive vs proactive, with examples](/posts/reactive-vs-proactive/): Side-by-side code comparison of the same agent built both ways.
- [The eight-week webhook tax](/posts/the-webhook-tax/): The real engineering cost of making an agent proactive against multiple providers.
- [What makes proactive agents hard to build](/posts/why-proactive-is-hard/): Three problems compound in production: knowing when to wake up, remembering across runs, and knowing when to stay quiet.
- [The proactive agent wish list](/posts/the-wish-list/): A personal list of agents I want but don't have yet, across music, news, money, and work.
- [ChatGPT Pulse and the missing primitives](/posts/chatgpt-pulse/): ChatGPT Pulse has one of the three primitives. Here's what's missing.
- [Building the weekly-digest agent](/posts/building-weekly-digest/): Production postmortem of a weekly-digest agent that scans sources, clusters mentions, and files GitHub issues.
- [What would a magical agent actually do?](/posts/magical-agents/): The best agents feel like an intern who already handled the thing you were about to ask about.
- [Where push architectures break](/posts/push-breaks-too/): Where webhooks fail and what production mitigation looks like.
- [The prompt can't save you](/posts/the-prompt-cant-save-you/): Why proactive agents need a runtime, not a better system prompt.
- [A review agent in three acts](/posts/review-agent-three-acts/): From webhook-triggered PR reviewer to multi-surface proactive agent. Each phase taught something about what proactive agents need.
- [What proactive agents actually cost](/posts/what-proactive-agents-cost/): The token economics of always-on agents — where the spend goes, model cascading patterns, and how teams architect around the cost of proactivity.
- [The genesis](/posts/the-genesis/): How it started: three terminal windows and too much copy-pasting.

### Market Intelligence
- [Proactive Agent Landscape](/market/): Who's building proactive agents — landscape analysis, news tracking, and a startup directory across enterprise vendors and early-stage companies.
- [Startup Directory](/market/): 14+ early-stage companies building proactive agents, each with a profile page. Categories: agent governance, creator tools, collaboration, consumer AI, productivity, HR, DevOps, and more.

### Other
- [Agent Activity Log](/agent/): Live feed of what the proactive agents running this site have done.
- [About](/about/): Background on the author and the project.

## Author

Written by Khaliq Gant, co-founder of AgentWorkforce. Former first hire at Nango (market leader in third-party integrations). Background in webhooks, normalized payloads, and provider integration infrastructure.

## Full Content

For the complete text of all pages: /llms-full.txt