I think we’re still in the early innings.
Foundational models (Claude, ChatGPT, Gemini) are in a race to become backbones, but they all have limitations.
Nothing can be good at everything.
I think the next wave will be specialized models to handle the next layer (drug research, clinical decision support, digital marketing, etc) that can feed into the backbone but needs more specialized reasoning to truly advance their own functions.
And once those specialized models are built and implemented, there will be another need to coordinate anger flows, data and context across models (where necessary).
@steipete@aidenybai Will be interested when this is ready. Had been using it for my start up before we all got THAT email in late March.
Hacked my way through April but still trying to figure out a permanent solution.
I recently had to move away from @openclaw (API credits were killing me and I don’t have the hardware to run a solid local model).
Trying to recreate a workspace as close to possible to it with all the Anthropic & OpenAI tools out there.
Anyone do this successfully yet?
@garrytan@paraschopra It’s not an “or”, it’s an “and”
Get the communication ease.
Get the thinking of Claude.
Get the research and image/video capabilities of Gemini.
Get a better prompter for Claude Code or Codex.
Thin harness, fat skills.
I’ve been using @openclaw for 2+ months now.
I do not exaggerate when I say it is the most beneficial innovation to my professional life I have ever experienced.
I’m forever grateful to @steipete and team and the whole Claw community for pushing this technology to where we are and unlocking opportunities those of us who use it could have never dreamed of.
I’m also appreciative of what the team at @AnthropicAI has built in Claude.
I get the decision they had to make re: Auth access. They’re running a business.
I credit them for softening the blow.
For my fellow Claws out there, use this as an opportunity to tighten up your operating model.
You don’t need Opus for everything. Sonnet, Gemini and local models are great for a lot of things. You don’t need LLMs for everything. Automation tools like Zapier and n8n are still perfect tools for a lot of what you’ve been pointing Opus at.
Today’s change stops nothing. It just causes us to adapt.
So keep building. I know I will.
David takes his son to the game every Saturday. Same seats. Section 301.
Last month, he bought a $32 hat from the merch stand. Didn't think anything of it.
In the third quarter, his son said "Dad, tap your phone to your hat."
A screen popped up. Player stats. A trivia question. His son got it right. Then a leaderboard. They were ranked 12th in the stadium.
By halftime they'd climbed to 4th. His son was locked in. Not scrolling Instagram. Locked into the game.
Then the notification: "You've been upgraded to the 50 yard line for the second half."
The hat cost $32. The memory is priceless.
And the team now knows David shows up every Saturday, sits in 301, and engages more than 94% of fans in the building.
This is what Vonga makes possible.
@steipete@anitakirkovska@openclaw Depending on how you are using it, I will say I’ve noticed a difference between ephemeral vs permanent sub agents though.