A look at custom chatbots, or GPTs, which are tailored for specific tasks and represent an important step in OpenAI's strategy of “gradual iterative deployment”
The age of autonomous A.I. assistants could have huge implications. — You could think of the recent history of A.I. chatbots as having two distinct phases.
New York TimesKevin Roose
Context & Ripple Effects
OpenAI’s custom-GPT builder was positioned as a deliberately simple first step, even as coverage flagged the risks of letting anyone create GPT agents. This makes task-specific bots a controlled way to test how people configure, share, and rely on assistants before they take on broader autonomy.
The arc quickly moved from creation to distribution: OpenAI later opened a GPT Store for publicly shared bots and reported millions of custom chatbots. Customization therefore matters not just as a product feature, but as a route to an ecosystem around reusable assistant behaviors.
First-order effects
Users and organizations can package instructions and task focus into purpose-built chatbots rather than repeatedly prompting a general assistant.
OpenAI gains a staged deployment channel for observing how customized assistants are built and used; the earlier focus on keeping the builder simple lowers the barrier to that experimentation.
Second-order effects
A public catalog creates incentives for builders to differentiate bots by usefulness and distribution, shifting competition from model access alone toward task design and audience reach.
As bots become more task-specific, expectations rise for guardrails and clearer boundaries on what a bot can do—especially given the identified risks of broadly available GPT agents.
Third-order effects
If customization continues to lead into device- and web-task automation, chatbots could evolve from conversational tools into an assistant layer that mediates routine digital work; reported agent work for devices and the web points in that direction.
The durable constraint will be calibrated trust: increasingly humanlike or autonomous interfaces can make users overestimate capability, so product design and governance may become as important as model quality.
The trend: AI companies are moving from one general-purpose chatbot toward distributed, task-specific assistants that can eventually act across users’ workflows.
@emollick Also, making RAG accessible without coding feels like a BFD, but it's too early to know for sure. I suspect we'll see a wave of GPTs for Q&A about specific topics / knowledge domains.
I don't know if you cats will be into this, but I made my own GPT, constructed from my 5,000 blog entries and several books I fed into the machine. You can upload any photo you like, and it will give you a critique with my experience and voice. https://chat.openai.com/... [image]
Roose quotes longtermism-funded CIAS on agents of doom & then concedes, “To be fair, OpenAI's bots aren't particularly dangerous.” Maybe he can make an agent that'll fall in love with him. Personalized A.I. Agents Are Here. Is the World Ready for Them? https://www.nytimes.com/...
@emollick There wasn't an Open AI prompt store before but now with the upcoming GPT Store, it will fuel a new app race that will be more no-code than ever before.
Having a GPT build another GPT is cool, but the real deal is a self-improving Assistant. My favorite Assistant hack is to make it self-improving by adding functions for updating it's own instructions or add long term memory as files for later retrieval. (dosen't work for GPTs... …
I think GPTs may have actually unlocked the potential of structured prompts for many people. I see tons of people sharing GPTs where before almost no one shared links to prompts. It is not really a capability change, but a user experience change. There is an old lesson there.
This is what I was looking forward to Instead of “here are 150 prompts” posts on LinkedIn At least now we'll have “here are 150 GPTs” which get you to value quicker I also love the public profiles of the builders, more accountability to deliver value is 👌
OpenAI is quite clever. By having people independently create “GPTs”, each with their own collection of uploaded knowledge and data, OpenAI is able to get around all of the pesky copyright restrictions that are currently holding back ChatGPT.
I thought it was a bit surprising that the uploaded files for the GPT RAG retrieval feature are also made available to Code Interpreter - looks like it's pretty easy to convince a GPT to give you a download link for that data