Workspace Agents in ChatGPT
openai.com
openai.com
The hardest part is ensuring that shared context is maintained and it converges on a representation of reality and the people in the company.
New knowledge additions are proposed when agents decide it would be relevant to retain, humans confirm/deny or create wiki modifications themselves.
It works, but models seem to have these insane long traces to do the most basic things. I had to create a couple of skills so they know how to properly use the thing without breaking, so they don't always try to pass the wrong parameters to it.
It also doesn't let us change a couple of things (like icons). Or, if it does, not even Opus 4.6 can figure out how to do it.
The main remaining part is the poor docx / pdf / final output but will create a skill/workflow to get around that.
Worked really well end-end!
"If only I had an agent which does everything at work for me". The logical continuation would be 'then I wouldn't have to work', right ?
No. Just as with coding agents, it doesn't mean less work - it means a lot more work of a different kind with the main challenge - don't loose your mind when managing the outputs from these agents.
Look at it another way - if these agents work perfectly and really increase productivity and profits and companies agentify all of their processes/development - then won't these companies essentially become extensions of OpenAI and not the other way around ?
If it helps, here's an example: Our team shares wonderful customer demos in Slack. In the past, these would get lost in Slack unless someone took the time to manually create the documentation and log it in Notion.
That meant no one really logged them and the example of good work was lost.
Now, I tag workspace agent in the slack thread, it reads the thread, puts it into the right shape, and logs it (with considerably high fidelity). Saves us time, does the job no one wanted to do, and helps new hires+tenured folks like me learn from our colleagues.
(I used to use Codex to do the same thing by sharing the slack link, but now can skip that step).
You may think its value add, but from what Im seeing from a valuation stand point theres no way the valuation of OAI makes any sense.
Agents at the edge of business where they can work independently, asynchronously, is an approach that I don't feel was explored enough in business environments.
Sending your entire communication and documents to OpenAI would be a very bold choice.
I do believe that LLMs and AI provide actual value, but the "workspace" is usually the passive aggressive CYA battleground for employees to appear productive in-spite of leadership's blind-spots, ossified business practices, and "aligned" decision-making that doesn't actually fix a broken org. Maybe this release will be the one that finally challenges nepo-hires, not-invented here, and all of the other corpo crap that defines "enterprise" business.
With this said, you can use your incorrect AI answers to find and then purge or repair this old and/or poorly written documentation and improve the output.
How many more are thinking “am I next?”
(I built https://nelly.is as a solo founder without funding)
So we’ve got about a year and a half max until we have AGI and OpenAI is launching a bunch of in house harnesses.
They must have some crazy shit cooking in the back rooms. So super duper top secret they can’t even announce it. Because if their public models are any hint, you would never think we were 18 months away from human level machine intelligence.
I'm so tired of seeing these companies trivializing other people's work! Nobody's job is "edit files" and "respond to messages"! People have jobs like "find and close leads" and "reconcile accounts" and "arrange student field trips" and "make sure the hospital has enough inventory", not "generate reports" and "write code".
Editing files, producing reports, even writing code is just a byproduct. This is like the idiotic "lines of code produced" metric, but now they apply it to all of society.
Yes, work is being trivialized, but the symptom here isn't caused by that.
But the problem is LLMs can't do those things. All they can do is "edit files" and "send messages".
The scarce resource preventing more people from the ideal solution of using a script in your scenario is you. Most people can’t write a script, so their options are slow and “expensive” manual process, or the 100x as efficient AI. The 1000x as efficient script isn’t an option (well, until the model is good enough to know it should obviously just write the script too).
- How you keep on top of what they are up to?
- How do they organize and coordinate?
I think this can only work based on a solid agent id system.
Shameless plug: I have been working on a solution for it, available at https://github.com/awebai/aweb and with a distributed, independently verifiable, and fully open id system at https://awid.ai
I wonder if this could be made to work with OpenAI's workspace agents.
another thing: this is all on OpenAI's servers. Which is fine if that's what you want. But there's a real class of user — technical, working on actual production code, security-conscious — for whom "my workspace lives on my machine, in my git repo, under my version control, works for my other non-openai tools" is a hard requirement, not a preference.
Edit: To answer my own question "Workspace agents are available in research preview in ChatGPT Business, Enterprise, Edu, and Teachers plans."
I'm either genuinely missing some key harness/whatever, or businesses are going to have issues down the line.
The AI agents coding up everything in parallel is just the latest iteration of a series of fads. Previously it was "outsource everything to low-cost Indian developers". Before that, it was "visual drag & drop rapid development". Before that, it was 4GL. Etc, etc...
Turns out that coding is more like a proof of understanding by the humans responsible.
You can't outsource responsibility.
Zzz this is boring. So much for scaling up compute and data = intelligence.
Funny we landed on the same terminology. Will need to connect Stripe.