The People Around You Already Using AI Aren't Using 'Chatbots'

A Scene That Made Me Pause for Three Seconds

Last month I had dinner with a friend who makes indie games. He casually mentioned what he's been using AI for — not writing scripts, not generating images. He's been using Claude to automatically compile his Steam reviews every week, categorize types of player feedback, and output a "this week's experience issues to fix" list.

He spent about half a day building the whole workflow. After that, he never had to think about it again.

My immediate reaction was: "That works?" Then two seconds later: "Yeah, that completely works."

That story made me realize something: the truly interesting, innovative uses of AI tools rarely come from some big announcement. They come from someone quietly wiring it into a place nobody thought of, and then things just start running smoother.

---

Most People Are Still Stuck in "Q&A Mode"

Think about how most people use AI today: they walk up to the microphone, ask a question, AI answers, they write it down, copy-paste, close the tab.

Nothing wrong with that — but it has no compounding effect. Every single time, you're starting from scratch: re-asking, re-organizing, re-deciding what's useful.

The people doing truly "innovative" things with AI have already moved past that layer. They've wired AI into their workflows — not as a thinking aid, but to directly execute a specific segment of work.

A few examples I've observed recently:

• Content creators: Not using AI to write articles, but using it to take things they've already said (podcast transcripts, interview recordings) and reorganize them into a library of materials in different formats, ready to draw from on demand
• Product designers: Using AI to turn user interview recordings into a "pain point tag list" that feeds directly into a design brief
• Small e-commerce businesses: Feeding customer service emails in and having AI generate a weekly "most common unresolved issues report" for human reps to tackle the highest-frequency pain points
• Engineers: This almost goes without saying — tools like Claude Code are already reshaping entire coding workflows, not just completing snippets but scaffolding entire features

What do they all have in common? They've all positioned AI at the "information comes in → gets organized → outputs a usable format" node, rather than "let me ask you a question."

---

A Workflow I Tried That Didn't Quite Work Out

I've tried wiring AI into my own workflows too. Not every attempt has gone smoothly.

For a while I experimented with having AI help me with "weekly reading digests" — dropping in article links I'd saved, asking it to extract key points, categorize topics, and generate summaries. Sounds great in theory. In practice, it broke down immediately: AI couldn't access the live content of those pages (the session had no browsing capability, or the information was stale), and the whole workflow collapsed at step one.

I eventually worked around it by using another tool to pull article content as plain text first, then feeding that in. It ran, but it added a manual step — and that's enough to take the edge off the compounding effect.

This pointed me to a practical reality: many AI workflow innovations don't stall because the AI isn't smart enough — they stall at "how does the data get in." If your input end requires manual preparation, automation isn't quite as automatic. There's no perfect fix, but being aware of it means you'll think one step further when designing your workflow.

While we're on the subject — this is also why AI accuracy remains a foundational risk. Do you know how often AI is wrong when you ask it questions every day? That's not a clickbait headline. It's something you need to understand before you wire AI into any automated process.

---

Who's Most Likely to Find These "Strange Connection Points"

From what I've observed, the people most likely to discover innovative AI applications aren't the most technically fluent — they're the ones who are most keenly aware of which repetitive tasks fill their days.

Think of it as reverse engineering: instead of asking "what can AI do," ask "which three things do I do every week that are purely mechanical — things I hate doing but can't avoid?" Then go find out if AI can absorb one of them.

The indie game developer knew he read reviews every week. The designer knew she was always re-organizing interview notes. The e-commerce owner knew his support inbox was full of the same questions on repeat.

Those pain points are where AI tools actually become valuable. Not in the flashy demos.

---

Something Worth Watching: Open Source + Local Deployment Is Changing the Game

There's another trend that deserves attention. More and more companies are asking: why am I feeding our internal data into someone else's SaaS? The core of this was articulated well by the Hugging Face CEO — once open-source models are good enough and can run locally, many use cases shift from "you can do this with ChatGPT" to "you can run this on your own machines, and the data never leaves."

For anyone looking to wire AI into their workflows, the significance of this shift is: privacy and control are no longer trade-offs against capability. You can have both.

---

One Thought to Take With You

Next time you're scrolling through coverage of new AI tools, try a different lens: instead of asking "what does this tool do," ask "where in my workflow could this tool plug in, and then run without me watching it?"

The integration that lets you "not have to think about it" — that's the real innovation. Everything else is probably just a better-looking chat interface.