AI is Getting TOO Smart: One Assistant or Many Tools?

Description

Chapters

Transcript

0:00 from coder dad. Between Google showing
0:02 off deeper Gemini integration across
0:04 search, Android and workspace and open
0:07 AI pushing codeex forward. Are we headed
0:10 toward one AI assistant that follows us
0:13 everywhere or are we or will we still
0:17 need different tools for different jobs?
0:19 Okay, great question. We are in the
0:21 middle of those two worlds. This is what
0:25 what was just described here. So there
0:27 they're
0:30 The last three and a half years, the AI
0:33 tools have largely been these discrete
0:37 things that we have to go learn, right?
0:39 You go to chat GPT, you learn how to do
0:42 custom settings, you learn how to do
0:44 custom GPTs, you learn how to um do
0:48 prompt engineering and and do structured
0:50 prompting and things like that. You you
0:52 talk into chat GPT. um you get out a
0:56 result like us as a human are going to
1:00 the tool using the tool and the tool is
1:03 responding and then we're doing
1:05 something with that you know we're
1:06 publishing it or we're putting it in
1:08 some other chain of craft. The new world
1:10 where we're headed is a world where
1:15 whether it's Codeex or OpenClaw or
1:18 Hermes or um or or the new Google
1:22 version of this, the AI the AI uh studio
1:25 version of this. They've got a consumer
1:28 version of it called Spark that they
1:30 talked about.
1:32 What these things do is they run 24/7.
1:38 um they can take action without you
1:42 um
1:44 needing to talk to them, right? You can
1:47 just give them a goal. In fact, one of
1:48 the features, one of the new features in
1:50 codeex is goal. I have a goal that you
1:54 will go do this thing for me at some
1:56 point somehow
1:59 and that's it. And then it will go off
2:02 and do stuff. Now, why I say that we're
2:04 in between these two worlds, all of the
2:07 discrete individual tools are getting
2:09 more and more and more complicated.
2:11 They're getting more and more
2:12 sophisticated,
2:14 and they still largely
2:17 suck. They still largely don't quite
2:21 give us what we need, right? You know,
2:23 Opus 4.6 was really good at coding and
2:26 then they launched Opus 4.7, which you
2:28 think is going to be better, but it's
2:29 actually not. they kind of borked the
2:32 model and they made it stupider and
2:36 and chat GPT is awesome and then
2:38 sometimes it's not. So the tools
2:40 themselves
2:42 are okay but they still require our
2:44 participation. All these agent
2:46 frameworks
2:48 are all brand new and they're all janky
2:51 and they are all broken and they they
2:53 sort of do remarkable things and then
2:55 they sometimes don't and they sometimes
2:57 do. And if you're not super geeky and
3:00 want to get under the hood and figure
3:01 out how to do things like skills and
3:04 skill management and all sorts of things
3:07 like that, they're just kind of a mess.
3:09 And even if you do all that stuff,
3:11 they're still kind of a mess. So, we're
3:13 in the middle. We're in the middle.
3:14 We're in the goopy the goopy transition
3:17 between a world of discrete AI tools and
3:21 these agentic truly agentic tools that
3:25 are going to do things on our behalf.
3:26 Where we're headed clearly
3:30 is we're just going to be able to say,
3:33 "I want to start a business." And the
3:35 agents will do their thing and go help
3:38 you start a business.
3:39 >> Watch the full replay at
3:41 community.alon.ai.