essay

Why do people use AI chat for everything?

Jeff Reynar
Why do people use AI chat for everything?

At this+that we’ve held a few beliefs strongly since generative AI took the world by storm. First, we’ve thought the predominance of AI chat was an interim step and that ultimately there’d be new AI-powered versions of key enterprise tools for email, messaging, task management and even specialized but frequent tasks like data analysis and managing customers in a CRM. We weren’t alone in this. Google built Gemini into Gmail and Docs, Microsoft built Copilot into Outlook, Word and Teams, and many others made the same bet. We thought occasional tasks might remain in AI chat because it’s time-consuming to learn a new product and easy to forget how to use it well without frequent use.

We believed this due to the benefits of decades of UI improvement (e.g. it’s much easier to send an email today than it was in 1990 thanks to Gmail), and people building vertical products for increasingly narrow niches (such as Expensify for expense reports). There was even the joke in enterprise SaaS circles that most enterprise SaaS companies were simply a use-case from Excel turned into a product. The consumer equivalent was that founders unbundled many successful products from Craigslist and gave them great UI.

Today there are products that offer AI features coupled with more AI-oriented UI, but we still see lots of people using Claude/ChatGPT/Gemini to manage their email, deal with Slack / Teams messages, handle tasks and do more. I even find myself wondering about my own use of Claude Code for writing tasks. I sometimes start in Google Docs and edit with Claude, other times I do the opposite. For large changes it makes sense to use Claude, but if I’m rewriting just one sentence, surely it’d be faster to crack open a word processor, but I rarely do.

Our second belief is that AI would become less centralized over time. When electricity arrived in factories it typically replaced a central steam engine that used belt drive to power equipment. The benefits of electricity were marginal until people realized that they could put small electric motors close to each machine and organize factories based on the logic of production rather than the layout of power transmission systems. We argued in October that a chat window off to the side is the AI version of that central power source. We expected AI to move to the application since it could then be specialized to the task at hand and would probably be both faster and cheaper. Maybe that will happen but we talk to more people who use their CRM today as a backend rather than using the AI integrated into it like HubSpot’s Breeze.

As a startup, and at the risk of sounding cliché, we need to “skate to where the puck is going” which encourages us to think about why our beliefs are wrong today and whether we’ll flip back to more specialized UI and local AI. That in turn leads us to wonder why people use Claude and ChatGPT for virtually everything today. We’ve looked for research about this topic but are underwhelmed, so we’ll throw out some hypotheses. We’re thinking out loud here. We may be wrong, about our original beliefs or about these hypotheses, and we haven’t settled what any of this means for our product. We suspect that there’s no single answer but that a constellation of factors encourages people to stick with their favorite AI tool for virtually everything.

  1. It’s similar to working with a teammate, especially to people managers. Those who manage are accustomed to asking people to perform tasks and, particularly if they manage junior colleagues, they’re used to providing guidance on not only what to do but how to do it. This is similar to asking an AI agent to do something for you. A 2024 paper makes the same comparison, saying the demands of working with generative AI “parallel those of a manager delegating tasks to a team.”
  2. It seems natural. You don’t have to learn a new interface to get work done, you can speak your native language to the AI.
  3. There’s no task-specificity. Gmail works one way, Outlook slightly differently (e.g. labels v. folders) and both have UI entirely different from, say, Figma. But, if you’re using an AI chat system, you can issue commands pertaining to each in more or less the same way. There’s less of a learning curve to climb. This applies both to the user interface and the underlying data model.
  4. There’s almost no window/tab switching. Once you’re in Claude, you tend to stay there. It sounds minor but extra clicks slow people down and they need to reorient themselves, especially if the applications they’re switching between are different or they need to open files. A 2022 study in HBR found people spent just under four hours a week getting their bearings after switching apps. You can ask Claude to send an email in one prompt and then in the next one ask it to schedule a meeting or update a price in a spreadsheet.
  5. Perception of speed. I’m not sure using AI is always faster for the tasks I use it for. But I have a hunch that it may seem faster, even when it’s not. This may relate to naturalness or the feeling of speed that comes from forward progress. I’m reminded of my own preference to drive a longer route to get somewhere than a shorter one that might be gridlocked, even if it’s faster overall. The evidence is thin but points the same way. In METR’s 2025 trial, experienced developers believed AI made them 20% faster when it made them 19% slower, though METR has since said those results are out of date. In a 2026 preprint comparing a chatbot with a dashboard, people reported less mental effort with the chatbot but weren’t faster overall. And in a small study of 12 people using older AI, editing by selecting and manipulating text directly took about half as long as doing it through chat.
  6. Encouragement. None of the tools I use frequently, like Gmail, Slack, Google Docs, VSCode or Figma, tell me I’m insightful or doing a good job, let alone occasionally calling me a genius. Flattery seems to work. In a Stanford and Carnegie Mellon study of 1,604 people seeking advice on interpersonal conflicts, participants rated sycophantic responses as higher quality, trusted the AI more and said they were more willing to use it again. Maybe I’d use Google Sheets more often if it told me my formulas were brilliant. Heck, I might take more money out of the ATM if it told me $100 was an excellent choice or that I was frugal.
  7. Cost. Pre-IPO Anthropic and OpenAI are probably undercharging for their AI chat products. Once they’re public, that may change and price hikes might encourage people to question their approach to work, especially if companies become even more AI-cost conscious.
  8. Cost, reason 2. Once you have a plan from Gemini, OpenAI or Anthropic you can use it for virtually anything. $20 gets you pretty far unless you’re a real power user. Buying lots of task-specific AI plans can add up to more than $20 pretty easily. Again, this may change as these companies rationalize prices.
  9. Context. The more you use one AI chat system, the more it records about you. Claude, ChatGPT and Gemini can remember past conversations, and each task you bring adds to what the next one can draw on: who your customers are and what you decided last week. Spread the same work across task-specific tools and each one mostly sees the data it holds, unless you connect them.

We’re curious about a couple things. First, do you think the era of AI chat agents is waning? And what about your own use? Do you expect to shift back to using more task-specific tools and products? Second, if you question the wisdom of your use of AI chat for some tasks, do you have a theory about why you use chat even though it might be slower, lead to worse results or somehow be inferior to using a product tailored to your task?