Reactive AI
Definition
Reactive AI is software that produces output only in response to a user prompt or invocation. ChatGPT, Claude, Gemini, and most in-app AI features are reactive: they wait. The model is powerful when invoked but generates zero value when ignored. The opposite is proactive AI, which initiates action without being asked.
What “reactive” means in practice
Reactive AI runs the same loop: user prompts, AI responds, AI goes idle. ChatGPT, Claude.ai, Perplexity, Gemini are the canonical examples. The user is the trigger; without invocation, nothing happens.
In-app AI features inside Notion, Gmail, Outlook, Slack, and most SaaS tools follow the same pattern. The AI is available when you press the button or type the slash command. It does no work in the background.
The strength of reactive AI
Reactive AI is the right pattern for many use cases:
- Open-ended thinking work (research, writing, analysis) where the user is the bottleneck and AI accelerates output
- One-shot questions that don’t recur (translation, explanation, code generation)
- Variable-context work where the AI can’t predict what you need
For these, reactive is more efficient than proactive. Proactive AI guessing at what you need wastes attention. Reactive AI responding precisely to what you asked is high-signal.
The weakness of reactive AI
Reactive AI does no work between invocations. If your bottleneck is recurring work that happens on a schedule — daily inbox triage, weekly reviews, follow-up tracking — reactive AI requires you to remember to invoke it. Tools that depend on user discipline tend to lose to tools that don’t.
This is the structural limitation that proactive AI exists to solve. A user who forgets to open ChatGPT for a week loses a week of value. A proactive assistant that runs overnight regardless of attention generates value continuously.
When reactive is the right choice
Three cases:
- The work is variable. You can’t pre-plan what you’ll ask for, so a system waiting on prompts is the right shape.
- The work is rare. A monthly forecasting question doesn’t justify a proactive monthly system.
- The trust gap is wide. For high-stakes actions (legal advice, code review of production systems) you want explicit human invocation rather than autonomous action.
How to combine reactive and proactive
The strongest 2026 setup pairs both. A proactive assistant (alfred_) handles recurring inbox, calendar, and task work. A reactive tool (ChatGPT, Claude) handles open-ended thinking work. The proactive system frees the time the reactive system uses well.
Where alfred_ fits
alfred_ is explicitly designed as the proactive layer, complementary to reactive AI. Email triage, drafting, brief generation, and follow-up tracking run autonomously overnight. When you want open-ended thinking help, you reach for ChatGPT or Claude. The two patterns serve different work modes.
What reactive AI isn’t
It isn’t bad AI — it’s a design pattern with specific strengths. It isn’t necessarily a chatbot — many in-app AI features are reactive without a chat interface. And it isn’t going away — even as proactive AI grows, reactive remains the right pattern for variable, exploratory, and one-shot work.