The internet is losing its mind over AutoGPT and AI "agents" right now. Every tech newsletter is breathlessly announcing that AI can now autonomously complete tasks without human supervision.
As a business leader, you should know: This hype has some truth, but it's mostly hype. Here's how I see down what's actually happening and what matters for your firm.
What Is AutoGPT?
AutoGPT is a prototype that lets GPT-4 set its own goals, break them into sub-tasks, and execute them by calling APIs and reading the results.
For example: "Get me the top five stories about AI regulation this week." AutoGPT would:
- Decide it needs to search the web
- Call a search API
- Read the results
- Realize it needs to filter by date
- Call the API again
- Compile a summary
- Return the result
All without a human in the loop. It sounds amazing.
What the Hype Gets Right
This concept is real. AI that can chain together multiple steps and tools is useful. It's a real step forward from "I ask ChatGPT a question."
For research and information gathering, this could genuinely save time.
What the Hype Gets Wrong
It's not reliable: AutoGPT often gets stuck in loops. It misunderstands API responses. It makes decisions that don't make sense. It needs a human watching it.
It's not a product yet: AutoGPT is a demonstration. It's running on researcher's laptops. It's not something you can deploy to your team.
The economics aren't there: Running GPT-4 for every step of a multi-step task is expensive. It's often cheaper to have someone do it manually.
It can't do anything requiring judgment: If you need an AI to complete a complex task that requires understanding your firm's context, preferences, and constraints, AutoGPT isn't the answer.
What This Actually Means for Professional Services
Don't wait for AutoGPT. Don't plan your 2023 AI strategy around it.
Intelligent agents might be useful for your firm eventually. But not today. And not in the way the hype suggests.
You can do more valuable things right now with tools that exist and are working well.
What You Should Actually Focus On
Assisted workflows: Human + AI where the human is in control. AI drafts, human reviews. AI summarizes, human verifies. This works today and creates real value.
Specialized agents: Instead of generic AutoGPT, specialized tools built for your specific business. This takes longer to build but creates more value.
Integration: Connecting AI into your existing systems and workflows so it's not a separate tool. This compounds the value.
When Agents Matter
True AI agents that work reliably will eventually be valuable for:
- Complex information gathering (research)
- Routine task automation (scheduling, data entry)
- Customer service automation (answering questions from a knowledge base)
But this is probably 18-24 months away, not months.
The Signal vs. Noise Test
When you see AI hype, ask yourself: "Can I use this today on real work?" If the answer is no, it's noise.
ChatGPT was a signal because you could use it immediately.
GPT-4 was a signal because it improved on something you're already using.
AutoGPT is interesting, but it's noise. Not because the concept isn't good, but because it's not ready.
What to Do Right Now
Ignore the AutoGPT hype. Keep building your AI competency in operations. Keep testing ChatGPT and GPT-4 on real work.
When true AI agents are ready, you'll be positioned to use them well. Right now, they'll just distract you from the actual opportunity.
Stay focused on signal, not hype.