In 2023, every SaaS product added a chatbox to the bottom right corner of their app. "Chat with your data!" was the pitch.
In 2026, we've realized a painful truth: users hate typing prompts into chatboxes. It requires too much cognitive load.
The shift: From explicit chat to implicit action.
The best AI features being built today are invisible. They don't wait for the user to ask a question. They anticipate the workflow.
Examples of Invisible AI:
- The CRM Summarizer: Instead of a user asking "What happened on the last call with Acme Corp?", the AI automatically triggers via webhook when a meeting ends, parses the transcript, updates the Salesforce fields, and drops a 3-bullet summary into Slack.
- The Code Reviewer: Instead of pasting code into a chat window to ask "is this good?", an AI agent lives in your CI/CD pipeline, reviews every Pull Request automatically, and leaves inline comments about specific performance bottlenecks.
- The Triage Agent: When a customer files a support ticket, an AI doesn't wait to be prompted. It instantly reads the ticket, queries the internal docs, drafts a response, and tags it with a priority level for the human agent.
The Rule for 2026:
If your AI feature requires the user to type a prompt to get value, you've built it wrong. AI should be a background worker that pushes value to the user proactively.
If you found this helpful, I write a weekly newsletter for AI builders covering deep dives like this, new models, and tools.
Join here: https://project-1960fbd1.doanything.app
Top comments (1)
The shift from "chat with AI" to "AI runs in the background" maps directly to something I keep seeing in coding workflows.
The most effective AI coding setups I've used don't involve chatting with the AI at all. You write a spec file (markdown, not a prompt), define acceptance criteria, and the AI runs against it — like a CI job, not a conversation. If it passes your tests, great. If not, it retries with the error context.
This is basically what you're describing: AI as infrastructure, not as an interface. The interesting consequence is that the skill shifts from "prompting" to "specification writing." Good prompts look a lot like good specs — and specs are something engineers have been writing (badly) for decades.
The chat UX isn't dying because it's bad — it's dying because it puts the human in a synchronous loop with the AI, which wastes the one resource AI is supposed to save: your attention.