Latest / Nonsense with cj little and Jeff Parker / Beyond the Chatbot: AI Finally Gets a Job

Transcript

Beyond the Chatbot: AI Finally Gets a Job · Nonsense with cj little and Jeff Parker

• 0:00
  Welcome to Nonsense Season 4, Episode 7, our big story on our
• 0:02
  second show of the week. I'm Jeff.
• 0:04
  Parker, I am C Jodel. This time we're looking at
• 0:05
  beyond the chatbot. Why?
• 0:07
  2026 is the year AI finally stopped talking and started
• 0:10
  doing up next. Wait, like doing what?
• 0:12
  Well, like I have AI put events in my calendar.
• 0:15
  That's pretty simple. OK, yeah, I want to do things
• 0:16
  like call cast of Vega, find what the wait time is for a
• 0:19
  table of four. I want to have some Mexican
• 0:21
  food. And mostly it doesn't do it.
• 0:24
  Mostly it says, yeah, I don't really do that.
• 0:26
  You should first ask it where can I go for good Mexican food?
• 0:29
  Start with that 20 going to Casa Vega.
• 0:32
  I would hate to rely sure on an AI's version of what good
• 0:35
  Mexican food? Is we should do a segment on
• 0:37
  this show where once a week we ask the AI where to go for lunch
• 0:40
  and just go wherever it says and see what happens?
• 0:42
  Sounds awful. We're.
• 0:43
  Having like lunch at Bob and Ell's transmission shop, like
• 0:46
  try the synchros, they're delicious.
• 0:48
  You're like, wait, what? I don't think that.
• 0:49
  I had synchros for breakfast. I had synchros for breakfast.
• 0:52
  They stay crunchy in milk. OK, how AI quietly moved from
• 0:56
  your screen to your bank account?
• 0:57
  Let's go. And then the AI just get sullen
• 0:59
  and it's like I, I this is it. This is all you got really.
• 1:02
  You've disciplined at the AI I. Was just in Sam Oldman's
• 1:04
  account. He had a lot more commas than
• 1:06
  you. What's going on in here?
• 1:07
  I'm going to go back to your screen if you don't mind.
• 1:09
  I'll show myself out. That's what the AI says I.
• 1:24
  In 2026, the tech world moved beyond generative AI, the
• 1:27
  chatbot era into the concierge style agentic era, where AI is
• 1:30
  transitioned from a poet to a practitioner.
• 1:32
  While 2024 was about talking, 2026 is about doing.
• 1:36
  We moved past the Clippy stage, where AI suggests a better word
• 1:38
  for an e-mail, and have now reached a gentic parody where
• 1:41
  the AI doesn't just do a single task, it's capable of an actual
• 1:44
  chain of thought. Did you?
• 1:46
  Refer to Clippy as a poet. You might want to rethink this
• 1:48
  copy. Of course, lines have been drawn
• 1:50
  between those embracing the action economy and those
• 1:52
  protecting legacy moats. Microsoft, often Salesforce,
• 1:55
  have gone all in, pivoting their entire stacks around agentic
• 1:58
  workflows that allow AI to book travel, manage supply chains and
• 2:01
  close sales without human hand holding.
• 2:03
  MasterCard has even launched an agent suite, essentially giving
• 2:06
  AI its own credit card to execute autonomous transactions.
• 2:09
  However, a human firewall has emerged.
• 2:11
  Men and TikTok are resisting deploying advanced bot blocking
• 2:14
  to prevent agents from bypassing their ad heavy interfaces.
• 2:17
  In healthcare and finance, giant legacy players like United
• 2:20
  Health and JP Morgan are moving slower, citing compliance
• 2:23
  paralysis. And in the legal nightmare of an
• 2:25
  autonomous agent making $1,000,000 error.
• 2:27
  The shift is fundamental autonomy over assistance.
• 2:30
  In 2026, the competitive edge belongs to those who stop
• 2:33
  treating AI as an encyclopedia and start treating it as a
• 2:36
  coworker. But both are heavy and old, so
• 2:38
  it's OK. Can we break this conversation
• 2:40
  into 3 little chunks? OK, The technology of action,
• 2:43
  how this whole thing works, what is, why it's different than
• 2:46
  generative AI? Yeah, exactly.
• 2:47
  Magic, The economic tug of war. Who benefits where?
• 2:50
  Who doesn't benefit? Why people want you to be able
• 2:53
  to use agentic AI to do things and why others are vehemently
• 2:56
  digging their heels in? No, I thought.
• 2:58
  It was an only fans page. Why would it be an only fans
• 2:59
  page? Economics.
• 3:00
  Tug of war. I'll let the listeners write in.
• 3:02
  OK, I know. And the new human operating
• 3:04
  model, which is humans as project leads.
• 3:07
  Yeah, As opposed to people just looking things up using
• 3:10
  generative AI, Sure. Or having funny pictures made.
• 3:12
  The other way to think about that is like, OK, what can
• 3:14
  humans do that the AI can't? Because you're running out of
• 3:16
  things to do if you don't approach it that way.
• 3:18
  Or your output is just being slowed tremendously because
• 3:20
  you're not tasking out things. Yeah, do the things that the the
• 3:23
  technical knowledge does well with the technology all.
• 3:25
  Right. Let's start with the technology
• 3:27
  of action. Things that generative AI did
• 3:30
  well, we still, we still need large language models.
• 3:32
  The whole thing is based on large language models.
• 3:34
  You can't get away from that. Well, yeah.
• 3:35
  World models are coming, but yes, right now that's what you
• 3:37
  got, yeah. I mean, we're not that far along
• 3:39
  in a gentic AI. We just started.
• 3:41
  No, but it's frustrating that I can't have it make a lot of
• 3:43
  calls and do things for me would.
• 3:45
  You rather do this show next month when it can do all those
• 3:47
  things, or the month after when it's taken over the world?
• 3:49
  How? When exactly would you like to
• 3:51
  do it 'cause it's all coming? I'm fine either way, I just want
• 3:54
  it to. To be able to find my luggage.
• 3:55
  You are ridiculous. It can do all of these things
• 3:58
  and you're like, you know, I just wish it could do a little
• 3:59
  more. It could make videos of Smurfs
• 4:02
  driving in cars. Like what?
• 4:03
  I don't understand. I want this thing to be able to
• 4:05
  make phone calls and be on hold for me and especially complex
• 4:08
  things like when I have to call 19 different people.
• 4:11
  What do you have to do in your life?
• 4:12
  We have to call 19 different people tracking down luggage.
• 4:15
  Have you ever tried to track down luggage?
• 4:17
  Let it go. What did nothing you wanted in
• 4:19
  there? The old laundry?
• 4:20
  Anyway, I don't. What's the problem?
• 4:22
  I know, but I want my. I have a tile in in there that
• 4:24
  tells me where a geographic tile bag.
• 4:26
  I want the tile that's. All you want you like I actually
• 4:28
  don't care about the bag. You could just take the tile out
• 4:30
  and send it back there's. Stuff in it.
• 4:31
  I don't care that man those. Things are like 12 bucks.
• 4:34
  So what am I, Rockefeller? Just send it back.
• 4:35
  It'd be great. Yeah, I understand.
• 4:37
  Yeah, I know you well. OK.
• 4:38
  Great, I can always get too many luggage.
• 4:39
  There's a hundred places that sell too many luggage, but those
• 4:41
  tiles are hard to get. Those tiles are hard to get.
• 4:44
  Yes, we are just the biggest orcs on the planet.
• 4:47
  I'm with you. I do think the the orchestration
• 4:50
  you're talking about right where the AI can now, pardon the
• 4:53
  expression here, orchestrate other agents to do things is
• 4:56
  coming. Yeah.
• 4:56
  But right now you have real agents that can do real things
• 4:59
  that we can embrace. By the way, you're tipping
• 5:01
  something off that we should talk about, which is the fact
• 5:03
  that, yeah, we're not looking at a single AI agent.
• 5:05
  Yes, you're going to have a lot of little AI agents that do
• 5:09
  simple tasks and can interact with each other.
• 5:11
  You. Think about it sort of at the
• 5:12
  task level. I think that's the right
• 5:13
  application for the agent, right?
• 5:15
  Not one agent that can do all of the things, right, the copy and
• 5:18
  right and create the image of both things.
• 5:19
  No, each one of those is like a different little micro worker.
• 5:22
  And we're getting to the place now, right, where you can ask an
• 5:24
  AI, hey, I want to achieve X, right?
• 5:26
  I want to find my luggage, right?
• 5:28
  And it's getting better at being able to put together the pieces,
• 5:30
  the other agents that it needs, the activities that it needs to
• 5:33
  deliver that I think about the different pieces that they
• 5:36
  quote, know the agents know a lot of this comes from how you,
• 5:40
  you share your context window with it.
• 5:42
  And I'm being very careful with the words that I picked.
• 5:43
  What you're willing to share with the AI?
• 5:45
  Yes. Yeah, what you're willing to
• 5:46
  give up, I have embraced for quite some time now.
• 5:48
  I was one of the first users of the Limitless AI PIN PIN PIN,
• 5:53
  which is really nothing more than a microphone phone that
• 5:55
  records everything that I say so that I can pump it through this
• 5:58
  system. Sends it to a server somewhere.
• 6:00
  Sends it off to a server so then they can get an even better
• 6:02
  profile. So that ice can really know
• 6:04
  what. You're so that context window is
• 6:07
  really important to establish. You don't have to do it with a
• 6:09
  with a pin that that tracks everything that you say, but you
• 6:11
  have to give it some context of what you're doing.
• 6:13
  The more specifics you can give it, yeah, the the better the
• 6:15
  answer you're going to get. And you were touching on this,
• 6:17
  Jeff, you know, in the wrong hands that context is is not
• 6:21
  good. But for you personally, that
• 6:24
  text is amazing because you have this.
• 6:25
  What we describe is the infinite context window.
• 6:27
  To have everything that you've said or done in a query able
• 6:30
  state to me is sort of the Holy Grail.
• 6:32
  Where it lives, who has access to it.
• 6:34
  Those are some real nuanced details that are important.
• 6:36
  And once you've established some portion of that context, now
• 6:38
  those agents can have virtually instant access to it.
• 6:42
  I mean, access that is measured typically in milliseconds, maybe
• 6:45
  single digit seconds. Yeah.
• 6:46
  But these are things that might take you 10s of hours to do.
• 6:48
  Right? Looking it up, maybe you wrote
• 6:49
  it down somewhere like it just has all of that.
• 6:51
  That's pretty impressive. So you have, you have the
• 6:52
  memory, you have sort of the ability to drive into that
• 6:54
  memory. You've got the ability to sort
• 6:56
  of plan and more importantly, adjust that plan when steps
• 6:59
  don't work. Sure.
• 7:00
  I still think we're, we're not as far along there as we'd like
• 7:02
  to see. That was your point earlier on
• 7:03
  the sort of the, the, the orchestration and not being able
• 7:05
  to figure out how to find my luggage.
• 7:06
  But yeah, it's getting better. And then being able to interface
• 7:09
  with the the the tools that you and I use as human.
• 7:12
  This is the big one, right? Yes, like you gave the example
• 7:14
  earlier, I know we kind of gloss over it, but I, I really do
• 7:16
  think is is the one of the building blocks of like putting
• 7:19
  something on your calendar or sending an e-mail or, you know,
• 7:21
  a text message, whatever the thing.
• 7:22
  That's fairly. Easy because the.
• 7:24
  Of course, I'm asking to put items on my calendar is made by
• 7:27
  the same company that makes my calendar.
• 7:29
  Yeah. So those things talk hand in
• 7:30
  glove all the time. An e-mail, same thing.
• 7:32
  Those things work great. It's when I want to have it talk
• 7:34
  to my bank. Sure.
• 7:35
  And my bank has no association with the company that's owns my
• 7:39
  AI. Then it's a a completely
• 7:40
  different issue. And for sure what I think needs
• 7:42
  to happen is obviously there are human interfaces for me to deal
• 7:45
  with my bank. I need like the ultimate
• 7:47
  skeleton key. I need the agents to be able to
• 7:49
  see and click screens just like humans can.
• 7:53
  Sure. So that they can do things with
• 7:54
  my banking. Yeah, You know, I think that's
• 7:56
  interesting. I'm I'm, I could see a world
• 7:58
  where that's the implementation. I could also see a world where
• 8:00
  it does these things, you know, in the in the shadows of the
• 8:04
  silicone, but presents you, the human with a like.
• 8:08
  Here's what I've been analyzed. Here's what I'm going to do.
• 8:10
  You now have to confirm this like it shows you the action.
• 8:12
  Plan well that's true, but I'm not worried about that so much
• 8:14
  as I'm worried about the bank saying no, we don't want oh AI
• 8:18
  understand you know coming and communicating with us and
• 8:20
  therefore the only way it can get in is through that web
• 8:23
  interface I'm probably. Really optimistic on this, but
• 8:25
  it seems like there's going to be two groups of of businesses
• 8:29
  out there, ones that embrace this and continue on and ones
• 8:31
  that don't and go away. To me, the real problem
• 8:34
  technologically is the the engine.
• 8:36
  The agents have outpaced by several orders of magnitude the
• 8:39
  interfaces. Yeah.
• 8:40
  And there's not a lot of motivation to build those
• 8:43
  interfaces right now. If you could also build the
• 8:45
  engine, meaning if you're a Frontier lab, go build the
• 8:47
  engine. That's where all of the money
• 8:49
  and interest is. And that's where you're making
• 8:50
  these huge game changing steps you sort of need.
• 8:53
  I hate saying this, but you almost need like the B team to
• 8:55
  come along behind it and be like, oh, we're building the
• 8:57
  interfaces because they're just not as interesting.
• 8:59
  If you can build the the core frontier model, sure, sure.
• 9:03
  And I do think that's happening, right?
• 9:04
  You're seeing things now. Again, Tropic released MCP a
• 9:06
  couple years ago. This is the model context
• 9:08
  protocol, which is not what you're describing, but is a big
• 9:11
  step in that direction, trying to get some sort of agreed upon
• 9:14
  quote UN quote standard. That's the problem is how do you
• 9:16
  get sign off? How do you get companies,
• 9:18
  especially companies who want you to come to their app totally
• 9:21
  and do your things through their apps so they have your data and
• 9:23
  they have your or experience and all of that in in the in hand?
• 9:26
  And I've seen some encouraging changes in this.
• 9:28
  Let me let me explain. Think about like, Wi-Fi
• 9:29
  standards. Those take years to be presented
• 9:32
  and ratified. Sure.
• 9:33
  Right. And that's something as, quote,
• 9:34
  innocuous as a Wi-Fi standard. It's really not innocuous.
• 9:36
  Like the things that they do, the problems that they solve,
• 9:38
  pretty cool. Yeah.
• 9:39
  Are pretty incredible. But it's on this time scale.
• 9:41
  It takes a long time, right? Like even from final draft to
• 9:44
  ratification can be 18 months or more.
• 9:46
  Yeah. So now you're in an AI world
• 9:47
  where like 18 months ago the AI thing you're trying to interface
• 9:50
  to didn't. Exist, but nobody has incentive
• 9:52
  to stop the next version of Wi-Fi from coming out.
• 9:55
  Think about all the companies who want you to use their app
• 9:58
  and not bypass their app. Of course, they have strong
• 10:00
  incentive to keep this. From happening, yeah, those
• 10:02
  folks are likely not going to be very participatory in finding
• 10:06
  the standard and they're also not going to be very
• 10:08
  participatory in implementing the standard once it's made.
• 10:10
  Yeah, I don't think they have a choice.
• 10:12
  They're betting on this not working at.
• 10:14
  All right, once there's somebody who does it exactly.
• 10:16
  Right now you're just being sand in the cogs and we've we've seen
• 10:19
  this time and time again. I agree with you.
• 10:21
  The tooling is relatively straightforward, but if you
• 10:23
  didn't have it, you would have way less value in the what the
• 10:26
  things can do now. So I'm glad that the tooling is
• 10:28
  coming along and the folks that are participating with it are
• 10:31
  participating with it. Yeah.
• 10:32
  Let's move on to the second section.
• 10:33
  We kind of are in it anyway, which is the economic tug of
• 10:35
  war. Yeah, The people who are looking
• 10:37
  forward to this and who are. Yeah, Microsoft has clearly
• 10:39
  moved whole hog into AI and using AI in your workflow.
• 10:43
  Yeah. And there are others who are
• 10:44
  just absolutely doing everything they can to slow this down.
• 10:47
  You know, it's interesting when I talk to folks that I will say
• 10:50
  are technologists maybe not leaning hard into the AI world
• 10:53
  or or maybe even ones that are starting to lean into it.
• 10:55
  They like to throw Microsoft into the bus of like, oh,
• 10:57
  Copilot, it's not great. And I'm like, you're taking the
• 11:00
  wrong view. If your view is you've never
• 11:02
  used AI before and you are a word, a Microsoft Word, a
• 11:05
  Microsoft Excel user, when you get Copilot turned on, it's like
• 11:08
  you just unlocked your superpower.
• 11:10
  It doesn't mean there's not better ones out there.
• 11:12
  But my point is augmenting the existing workflows that people
• 11:15
  have is a hard thing to do. Sure, I think Copilot does a
• 11:17
  pretty good. Job of that, I think the more
• 11:19
  interesting models are like cursor where you can pick what
• 11:21
  the AI back end is that you want.
• 11:23
  For sure. That's way more interesting to
• 11:25
  me. Of course, and as a nerd, a
• 11:27
  fellow nerd, I agree with you on that.
• 11:29
  But I'm thinking about the non nerds right, the folks that the
• 11:32
  trope is the jobs that are eaten by AI right.
• 11:35
  The people whose job is quote gonna get eaten.
• 11:37
  They're gonna let them would be more likely, I think to to
• 11:38
  onboard with copilot. Maybe they grow into a well, you
• 11:41
  know what, I prefer a Claude over copilot.
• 11:43
  Sure, sure, at some point, but right now they're not on the tip
• 11:46
  of the spear trying to figure out how to augment their
• 11:48
  workflow with Claude or. There's subtle distinctions
• 11:51
  between those back ends that you eventually get very passionate
• 11:53
  about. Yeah, but you certainly don't
• 11:55
  start out passionate. About it, If Copilot is the
• 11:57
  gateway drug to get folks onto to a Genting style thinking,
• 12:01
  great, I'm all for it. I think that's lovely.
• 12:03
  Yeah. And think about a world where
• 12:04
  right now where if Microsoft didn't do any of this, they're
• 12:06
  just like, Nope, we make word processors and spreadsheets and
• 12:08
  you'd be like, you would not be a trillion dollar company.
• 12:11
  Right. No.
• 12:11
  Sure, sure, sure you'd be. Headed in the wrong direction.
• 12:13
  We sell copies of Windows. Well, no, you really don't.
• 12:15
  You don't. Not your business anymore.
• 12:16
  Yeah, you sure don't. We've talked about this part
• 12:18
  before on the show thinking about the the economics of this.
• 12:21
  Most businesses, I mean, I'd say damn near all of them, they
• 12:24
  unfortunately value the work output that people produce based
• 12:28
  on the time spent to do it, which I've always said for 20
• 12:31
  years that's going away. I've said it's a terrible
• 12:32
  metric. That's not the right metric.
• 12:34
  I'm not going to claim I know what the right metric is in all
• 12:36
  industries, but like it's not time in.
• 12:38
  It's, it's really work output. You have to know how you divine
• 12:40
  work out. Completing the task.
• 12:41
  Yeah, with AI, Oh no, like that all goes away.
• 12:45
  The job I would normally give to the intern or the new, the
• 12:49
  junior higher, it's like relatively unskilled and you go
• 12:51
  do some research or whatever and it would take them 30 hours and
• 12:53
  they produce, it's a six page PDF and I'd cost me, you know,
• 12:56
  whatever I, I pay them 40 bucks an hour.
• 12:58
  So it would. Cost now done in 12 seconds,
• 13:00
  yeah. For like $0.10.
• 13:02
  Yeah. And you're just like, oh, wait a
• 13:03
  minute, Like that junior roll now I don't need it.
• 13:05
  It's good and it's good. Yeah, it's good.
• 13:07
  It's not good. It's better than what I would
• 13:08
  have got by paying the junior. Sure.
• 13:10
  You know, the junior isn't up to that skill set yet.
• 13:12
  Like that whole stripe just goes away like on the org chart.
• 13:15
  You just don't need it anymore. Yeah.
• 13:17
  And a lot of people go all this terrible.
• 13:19
  I'm like, no, this is actually great.
• 13:20
  This is advancement. Now we gotta figure out how do
• 13:22
  you get into that stream, because I have an unfair
• 13:24
  advantage of 25 or 30 years of experience and they don't right.
• 13:28
  So they need to get their experience amount they're going
• 13:30
  to right. They're not all gonna just play
• 13:32
  Fortnite and sit at home. They're gonna get that
• 13:34
  experience. We just gotta figure it out.
• 13:35
  I'm. Actually a lot less worried
• 13:37
  about the next generation that I am about the past generation,
• 13:40
  about people in my generation. You and I are aligned 100% the
• 13:43
  people in my. Child's level seem to have a
• 13:45
  much more intuitive grasp of all this and they got there before I
• 13:49
  did. Yeah, my kid was programming
• 13:51
  using AI long before I got. He's way, way ahead of me on it.
• 13:55
  Yeah. So the next generation worries
• 13:56
  me less than the transition generation.
• 13:58
  The generation, well, we used to do this by hand, and now there's
• 14:01
  nothing by hand here to do. Totally.
• 14:03
  We've mentioned this before on the show too.
• 14:05
  This is sort of in the the economics of this.
• 14:07
  Yeah. I am now completely come around
• 14:10
  to the idea that the current pricing model that we have for
• 14:14
  I'll just call it, you know technology, software, whatever
• 14:16
  the service based model is dead. Gone.
• 14:18
  It's over software. Service is over 100%.
• 14:20
  And what's really fascinating to me is it's going to be outcome
• 14:23
  driven. Like you're going to pay for the
• 14:24
  outcomes, which is brilliant. Like, I think that's amazing for
• 14:28
  businesses. This is like the shift we saw
• 14:29
  from selling ads on eyeballs, right, Which was the paper
• 14:33
  model. Like I print a newspaper.
• 14:34
  Yeah, and I have distribution and it goes to 300,000 people.
• 14:37
  I'm gonna charge you for the ad based on the number of
• 14:39
  newspapers that I sell. That was the model when we
• 14:41
  started the Internet. We did the same thing, 100,000
• 14:44
  people. See.
• 14:44
  That's what advertisers knew. But pretty quickly, a handful of
• 14:46
  folks, of which I was one, said no, how about we, we do it a
• 14:51
  little differently. How about we say no, we're gonna
• 14:53
  sell for people that actually respond to the ad.
• 14:55
  Yeah, Right. And do something with the ad.
• 14:56
  Yeah. And that was a big shift.
• 14:57
  Now we take it for granted now, right Now, you pay for clicks,
• 15:00
  you pay for acquisitions now, Right.
• 15:01
  That's just sort of the way that it works.
• 15:02
  Right. And I see the same thing
• 15:04
  happening to these like, well, I couldn't.
• 15:06
  I don't know how you're gonna use the CRM.
• 15:07
  So you just pay for it per seat per month.
• 15:08
  Over. Absolutely over.
• 15:10
  Yeah, of course. What's the new metric?
• 15:11
  Like, how can you describe exactly the outcome?
• 15:13
  I have no idea, but I know that it's it's coming.
• 15:15
  And it's gonna be probably on a per business or per product
• 15:17
  basis. And it's gonna make a lot of for
• 15:19
  the customers. I think it's a great thing go.
• 15:21
  Back to what you're talking about about advertising for just
• 15:23
  one second there. Sure, because companies like
• 15:25
  Meta and TikTok are just really slowing down every piece of
• 15:30
  progress because they want eyeballs for their ads.
• 15:33
  That's their model. What do they do?
• 15:34
  The future of advertising is what I want to talk about here.
• 15:36
  The future of advertising is AI can make a custom ad that is
• 15:41
  designed just for you and because on your TV as you're
• 15:44
  watching your Netflix, we know yes oh, this is CJ watching this
• 15:47
  show. We can now cater not just the
• 15:49
  product we're selling to CJ, yes, but how we sell it to.
• 15:52
  Him. I mean what?
• 15:52
  Kinds of things does CJ find entertaining?
• 15:54
  Let's make our ad for these Milwaukee tools so that it
• 15:58
  really appeals to CJ uniquely, just to that one person. 20
• 16:02
  years ago, we call that performance based advertising,
• 16:05
  right? We would say, yeah, let's show
• 16:06
  ads to Jeff, right, that he responds to.
• 16:09
  And we look at the other ads that you've responded to and we
• 16:11
  say, well, we think you're likely to do this.
• 16:12
  And that made the advertising world better.
• 16:14
  One of the few things I think they made it better for the
• 16:16
  consumer. I think advertising is going to
• 16:18
  get more entertaining, not less entertaining.
• 16:20
  It's going to get more engaging, not less.
• 16:22
  I think it might become part of the effing entertainment.
• 16:24
  Yeah, there's, I don't want to say an infinite, but man, it
• 16:26
  feels like an almost infinite amount of places This could
• 16:30
  work. Like you could even see
• 16:31
  something where like the ad placement sure inside the
• 16:34
  content is custom to you, it'll get.
• 16:36
  More and more bespoke and if you.
• 16:37
  Want to sell me shoes? And you've got some set of data
• 16:40
  that says, oh, we think you're really going to like these
• 16:42
  Adidas shoes. Now instead of me seeing a 32nd
• 16:45
  advert that I'm going to go pop popcorn while that's playing,
• 16:48
  instead, the main character in the show is wearing the Adidas.
• 16:52
  You want to sell me the. Shoes you were looking at
• 16:54
  online. Oh, that's actually, I think
• 16:56
  pretty cool. Like it's a little creepy, but I
• 16:57
  think it's pretty cool. I guess what I'm if I, if I zoom
• 16:59
  out a little bit, the ad world has to change.
• 17:01
  Oh yeah, yeah. But as it's constantly done, it
• 17:04
  has to. I'm amazed that the ad world
• 17:05
  hasn't changed more yet. We, I mean, especially with this
• 17:08
  with streaming video Trench, it's amazing if you see ads on
• 17:12
  streaming video or how come those aren't talking to your
• 17:14
  Amazon or your Google experiences and saying, I know
• 17:18
  Jeff wants to buy yeah, a new pool cleaner.
• 17:21
  Yeah, why don't I put an ad up on his his Netflix show that
• 17:24
  shows him these new pool cleaners?
• 17:26
  That particular example you gave, I think the primary answer
• 17:29
  there is walled gardens, because now you need Amazon Netflix to
• 17:32
  play well together. But that doesn't answer the
• 17:34
  question of why don't you see? That that's just a matter of
• 17:36
  writing a check, but that. Doesn't answer the question of
• 17:37
  why you don't see those ads on Prime because then there is no
• 17:40
  walled garden. It's the same thing, right?
• 17:41
  Yeah, that's like convenience factor though.
• 17:43
  I think that I think there's more interesting stuff here and
• 17:45
  this is what we're we're starting to get into.
• 17:47
  I think around how humans interface.
• 17:48
  I'm thinking especially for work product, I don't think anyone is
• 17:51
  worried about your TV getting smarter and that ruining their
• 17:53
  career. No, sure.
• 17:54
  But getting back to like the work and how is this going to
• 17:57
  affect what I'm doing? I think that's that's
• 17:59
  interesting. And I think you were calling
• 18:00
  that earlier the the human OS, the human operating model?
• 18:03
  That by that's our third section.
• 18:04
  And by the way, let's move on to the third section, the new human
• 18:07
  operating model. Like new humans, we're building
• 18:09
  new humans. Where do you fit into this new
• 18:11
  agentic AI world? Well, I'm at the bottom of the
• 18:13
  of the pole. I'm I'm just at the very bottom
• 18:16
  doing my best to not get pushed underground.
• 18:18
  Let's talk about yes, to move forward in the world in yeah, in
• 18:21
  careers, you really need to be a manager, A-Team.
• 18:24
  I think that's the even though, yes, your team is all robotic,
• 18:28
  doesn't matter. Your teamer is all agentic AI,
• 18:30
  it does, right? You still have to know how to
• 18:31
  manage them. How do I get them to book the
• 18:33
  airline ticket? It's for the conference that I
• 18:35
  need to go to, book me a hotel, give me all that information,
• 18:38
  and if there's something I have to make changes or whatever, how
• 18:41
  do I oversee it so that it's not using my credit card to spend
• 18:43
  money on things that I don't need or want?
• 18:45
  Yeah. How do you make all that stuff
• 18:46
  work? How do you manage those that
• 18:48
  that? Workflow, I mean, this is where
• 18:50
  I think you still have to treat that AI like a tool.
• 18:52
  Yeah, but it's more than that. You want to interface with it
• 18:54
  like a person, but you have to remember that it's execution's
• 18:57
  gonna be very narrow, right? More tool focused.
• 18:59
  At the end of the day, it's really not either of those.
• 19:02
  It's right in between. It's like on the seam in between
• 19:04
  those two. Things I wanted to be able to do
• 19:06
  a three or four steps, though I don't want it to check back with
• 19:09
  me and just every single time I have to tell it what the next
• 19:11
  step is. I and and we might not quite be
• 19:14
  there yet. I find my luggage example.
• 19:16
  But I think we're very close. Well the find your.
• 19:18
  Luggage. That's so many steps and I don't
• 19:20
  want to know about any of the details of it.
• 19:22
  Just give my luggage to me. I don't think we're there yet
• 19:24
  'cause you're orchestrating a lot of things that are out of
• 19:26
  band from each other. Yeah, multi step within the same
• 19:29
  narrow domain. We're there that we can do now.
• 19:32
  It's when you get into multi domain and you've got these,
• 19:34
  these multiple domains, you want to be expert.
• 19:35
  And I'm like, I'm not saying we're not gonna get there.
• 19:37
  I, I, I really believe we're gonna get there sooner than you
• 19:40
  and I are probably willing to give credit to, but it's just
• 19:42
  not there yet. I'm just going back to sort of
• 19:43
  how the how does the human operate.
• 19:45
  You want to think of this thing as I, in my opinion, a narrow
• 19:47
  tool, but you still can interact with it like it's a person
• 19:50
  sitting in the cube next to you. Yeah, I've seen a couple
• 19:52
  different ways to describe this. One of my favorite is you think
• 19:56
  about it as a Co worker, but it's a Co worker with some
• 19:58
  additional interesting facets. It never gets tired, right?
• 20:01
  Never needs a break, never needs to take a nap.
• 20:03
  Yeah, sure it will. Occasionally lie with complete
• 20:05
  confidence. Like, yes, don't, don't, but
• 20:08
  that's important. Don't.
• 20:09
  That's true. Think as if you think about the
• 20:10
  coworker paradigm. You think, well, I've worked
• 20:12
  with this person now for six months.
• 20:14
  They're not gonna. Lie.
• 20:14
  And mostly they're telling me the truth.
• 20:16
  I haven't I haven't found anything they've said to be
• 20:17
  faulty, so that's not the case. Yeah, the.
• 20:19
  AI gives 0.0 FS about that. You can never let your guard
• 20:23
  down. Every interaction is a new
• 20:24
  interaction, right? It's like playing roulette.
• 20:26
  Everyone is an independent outcome.
• 20:27
  It doesn't matter that blacks come up five times in a row.
• 20:30
  It could just as easily be read the next, the next rule or
• 20:32
  black. Yeah, I think that's really
• 20:33
  important for and then I think another, another challenge that
• 20:36
  we make that we've got to got to sort of fix in this operating
• 20:39
  model is when you have a Co worker that is decent, which I
• 20:42
  know is a is a big if yes, you typically can be more vague with
• 20:45
  your instructions. You can be a little more, yes,
• 20:47
  abstract because they already have the knowledge and you know
• 20:50
  they have the knowledge looser. Yeah, right.
• 20:51
  I do not think you can make that presumption yet.
• 20:54
  I think we'll get there as you get a bigger context window.
• 20:56
  But right now the more specific you are, the better outcome
• 21:00
  you're going to get. So right, you really want to be
• 21:02
  specific. I think that's on the input
• 21:03
  side. Those three things the most
• 21:04
  important. And then this should just be
• 21:06
  obvious 'cause you've already said it now 3 or 4 times.
• 21:08
  Please check the output like. Please make sure that what comes
• 21:11
  out. Oh, you're saying to the humans?
• 21:12
  The humans? The human for sure.
• 21:14
  Yeah, for sure. Don't just take the output and.
• 21:15
  Before you book the plane tickets, give me all the
• 21:17
  information about them and let me know everything about them so
• 21:20
  that I can. Yes, OK.
• 21:21
  What you're doing before you spend money.
• 21:22
  And I feel like that point and I might be stretching a little bit
• 21:25
  here, but I want to get your thoughts on this.
• 21:27
  I feel like that point makes the case for at least for, for the,
• 21:30
  the short term, maybe medium term generalists are going to be
• 21:33
  able to do better because a generalist can look at these
• 21:36
  outputs and be like, that passes the muster.
• 21:38
  I think that's OK. If you are a very narrow
• 21:40
  specialist and you just treat what comes out of the AI as
• 21:43
  like, well, this must be right. The AI did it.
• 21:45
  I don't know anything about it. It's smarter than me.
• 21:46
  Great. Well.
• 21:47
  That's not a specialist if you're deferring no, no, no
• 21:49
  superior knowledge. But but for something that
• 21:51
  you're not a specialist, right? If you're already a specialist
• 21:53
  in one bit of knowledge and you're trying to do other things
• 21:55
  because you've got a team now, yeah, it's like, oh, man, you've
• 21:58
  got to get at least a medium level of knowledge in that space
• 22:01
  so that you can do the sniff test when it comes out.
• 22:03
  I don't want to blow up this this conversation here we do a
• 22:05
  whole show on generalist versus specialists, sure, in in the AI
• 22:08
  world. But my, my, the only point I
• 22:09
  want to make is you, you really need to understand the output
• 22:12
  and make sure that it passes some smell test.
• 22:13
  And I feel like this is one of the bigger gaps that most people
• 22:16
  just don't do. They just take the output like,
• 22:17
  oh, I'm done here you go, right. I'm like, did you even read
• 22:20
  this? Your AI can book you on a
• 22:21
  private plane that goes around the world three times.
• 22:24
  So. That you can go from Los Angeles
• 22:25
  to San Francisco. Yeah, but in the prompt you said
• 22:27
  make it fun. Yeah.
• 22:28
  And you're like, I was just thinking up front in Southwest,
• 22:30
  not on JSX world. Trip, what are you doing?
• 22:32
  You are still gonna have to oversee.
• 22:33
  You are still gonna have to. To manage that part's not going
• 22:36
  away at some point, maybe in the near term.
• 22:38
  I don't think we're there yet. Yeah.
• 22:39
  And I think this is kind of fascinating in a good and a bad
• 22:42
  way. The good way is I see an
• 22:44
  explosion in work output, of course, in the human operating
• 22:47
  model. If I think about work output
• 22:48
  from the human operator, I see an explosion in what they can
• 22:51
  do. By the way, does this mean we
• 22:52
  need fewer workers? That's no way.
• 22:54
  That just means we get to do more stuff.
• 22:56
  Well, OK, that's what I'm getting to.
• 22:57
  Like I'm not exaggerating when I say there could be a one, maybe
• 23:01
  even 2 orders of magnitude and output that we're going to see
• 23:04
  in the next couple of years from a single human.
• 23:06
  Yeah, but I think one order of magnitude is pretty much a
• 23:07
  given. Like you're gonna be in right
• 23:08
  five to 10X no problem. OK, You just hit the, the, I
• 23:11
  think the real question on the head.
• 23:13
  What do we want to do with that? Do you want less workers?
• 23:15
  Do you want more work? Output No.
• 23:16
  I want a funner world. I want a better.
• 23:18
  Work, right. Also what does the world look
• 23:20
  like if you go well, OK, if we actually could measure our work
• 23:23
  output and we knew that with our hundred person company, this is
• 23:26
  how much work output we got every month.
• 23:27
  Now with AI we can do 10X that with the same sort of time input
• 23:31
  and the same cost structure or but we have to find out if 10X.
• 23:34
  The output or 100 X the output, because more likely it's going
• 23:37
  to be, you know, 1000 X the output.
• 23:38
  Is that something that is useful for our market, Yeah.
• 23:40
  Or are we better doing different things with that that effort?
• 23:44
  Yeah, let's make an assumption that the output is linear.
• 23:46
  Just just just for the sake of this, this, yeah, this argument.
• 23:49
  Because then I go, well, you know what would be kind of
• 23:50
  interesting? Maybe if instead of us just
• 23:53
  trying to grind ourselves down to the bone until we hit some
• 23:56
  magic age when we then all say we're retired, what if we just
• 23:59
  did a little less with our time for quote work and a little more
• 24:03
  with our time for something else, like reading?
• 24:05
  Or art. I've heard this my whole life.
• 24:07
  I know it never happened. It never happened, never comes
• 24:10
  true. We just want to do more.
• 24:12
  We just want to make everyone's quality of life well, or at
• 24:14
  least our quality of life better.
• 24:16
  Better I I still think that is, we see a lot of that here.
• 24:19
  It's a very American sort of ideology.
• 24:21
  If you go talk to folks you know, that are, that are in
• 24:23
  Spain that are Spanish, they're like, I'm taking 90 minutes off
• 24:27
  after lunch because it's what I do.
• 24:28
  Yeah. They don't work any less hard.
• 24:30
  It's part of their culture. You're talking about taking a
• 24:32
  siesta in the middle of the day? Yeah.
• 24:33
  You're coming back with a stronger brain than I'm coming
• 24:35
  back with after I have lunch for 60 minutes and I come back, Yes.
• 24:38
  Exactly. It's effective.
• 24:40
  And I've had this argument many times with folks where they're
• 24:42
  just like, oh, you're taking a nap.
• 24:44
  And I'm like, you know, much better.
• 24:45
  I operate after a nap. Yeah, like you'd rather have me
• 24:47
  just pounding coffee to try to like, that's idiotic.
• 24:50
  I'm with you. I think it's a little a little
• 24:52
  grandiose to assume that we're all automatically you get a 20
• 24:55
  hour work week. You're not getting a 20 hour
• 24:56
  work week, you're getting a 40 hour work week because that's
• 24:58
  the maximum amount we can make you do.
• 25:00
  It's our socially acceptable limit.
• 25:02
  So the only reason you're you're stopping, the only reason you're
• 25:04
  not your work weeks not getting more is because we have the
• 25:06
  social construct. Yeah.
• 25:07
  Now when you have AI, we just want more output.
• 25:10
  We just want a better world. We've sort of talked, I've
• 25:12
  talked a little bit about how I think about these coworkers and
• 25:15
  I think it's important to remember that that empirical
• 25:18
  experience, right? It's one of my favorite quotes
• 25:19
  in the world, which is that good judgment comes from experience.
• 25:21
  Experience comes from that judgment.
• 25:23
  You don't have that, at least not yet in our agents.
• 25:25
  In this new. Coworker it's not there right it
• 25:27
  doesn't have it lacks that lived experience and the the true
• 25:31
  intuition the gut feeling that you have whether or not you
• 25:33
  listen to it or not consciously it doesn't have that and while
• 25:36
  maybe and I'm gonna put a under score on maybe these agents can
• 25:40
  simulate empathy it really can't feel the stake of the project it
• 25:43
  doesn't own the. Project It has nothing at stake.
• 25:45
  Yes. So at the end of the day, you
• 25:47
  can make your agent accountable to do something.
• 25:49
  You can be like, I want you to go, you know, do some research
• 25:51
  on blah, blah, blah, or book a right, but the responsibility
• 25:54
  still lives with you. Yeah, I think that's really
• 25:56
  important and gets missed on these.
• 25:57
  Things right? No, the problem isn't with the
• 25:59
  tools, the problem is with the people mad at you.
• 26:01
  Yeah, exactly. So anyway, that's my that my
• 26:03
  sort of big take away from from this and a fine take.
• 26:05
  Away it is all right. We have to get out of here.
• 26:07
  But quickly before we do, have you seen or read anything good
• 26:09
  this last week? Well, yeah, if today's.
• 26:11
  Episode hasn't sufficiently freaked you out about the future
• 26:14
  of work. I would recommend that folks
• 26:16
  read Co Intelligence Living and working with AI by Ethan Malik.
• 26:19
  Oh, yeah, I'm not all the way through it, but I think it's a
• 26:21
  very interesting book. So this book was written in
• 26:23
  2024, mind you, So it's ancient by AI standards, Sure.
• 26:26
  But even now, most people are still using AI sort of like a,
• 26:29
  like a supercharged Google search bar, right.
• 26:31
  That has, like, a mild personality disorder.
• 26:33
  Sure. And Malika explains why you
• 26:35
  should be treating it like a. And I think this is his quote is
• 26:37
  like a brilliant but socially awkward intern.
• 26:40
  Yeah. Which I think is really the
• 26:41
  right way to think of it. He introduced the idea of the
• 26:43
  jagged frontier, where an AI can solve a complex coding problem
• 26:47
  in seconds but might fail at a third grade word puzzle.
• 26:50
  Sure. Which is something that if you
• 26:51
  had a really good software engineer, you'd expect them to
• 26:54
  be able to do both. But in this world, that's not
• 26:56
  the case. So you have to change your
• 26:58
  expectations. The point that he makes is that
• 26:59
  if you want to actually survive the shift from doing the work to
• 27:02
  directing the work. And I'd argue, and if you want
• 27:04
  to thrive in this new world, I think his book is the closest
• 27:07
  thing to a manual, at least that I've come across.
• 27:09
  Yeah, it's practical. It is slightly terrifying.
• 27:12
  Let me just under score that again.
• 27:13
  But I think it'll make you the smartest person in the room for
• 27:15
  your next AI note taker driven sync meeting that you have on on
• 27:19
  Zoom next week. So can I not go?
• 27:21
  Can I send my AI? I know you're not going.
• 27:23
  I don't even know we have a little We know we have an old
• 27:25
  school puppet there for you. It's great.
• 27:26
  Perfect Co intelligence living and working with AI available on
• 27:29
  all the usual suspects, Amazon, Audible and on Libby.
• 27:32
  It's on Libby with with your bubble library.
• 27:34
  Oh yeah, yeah, yeah. Anyway, how about?
• 27:35
  You I'm gonna go with the 1968 film 2001 a space odyssey
• 27:39
  because nobody understood AI better than the people in 1968
• 27:42
  man's first tool was a bone his last will be an agent. 2001 A
• 27:46
  Space Odyssey is the ultimate cinematic trip through human
• 27:48
  evolution sparked by mysterious alien monoliths.
• 27:51
  It begins with prehistoric apes discovering violence via tools
• 27:54
  and leaps to the future where tools have become sentient.
• 27:57
  Aboard Discovery One, the mission to Jupiter turns into a
• 28:00
  cold blooded chess match against Hal. 9000 Yeah, an unbeatable AI
• 28:04
  that prioritizes the mission over human life.
• 28:06
  But after a terrifying battle of wits, Dave Bowman disconnects
• 28:10
  Hal's brain and plunges into a psychedelic Stargate.
• 28:13
  He transcends time, space, and biology, reborn as the star
• 28:16
  child. Kubrick's masterpiece isn't just
• 28:18
  a movie, it's a haunting prophecy about the moment our
• 28:21
  creation stop being assistance and start becoming our
• 28:23
  replacements. If you haven't seen it, check
• 28:25
  out 2001 A Space Odyssey on HB OS Max right now it.
• 28:29
  Has been a hot minute since I've seen this film, but I definitely
• 28:32
  go watch. Oh, it's a fantastic film and so
• 28:34
  many great lines And like also, I mean, I know you're throwing a
• 28:37
  bit of shade about 1968 being the the prophecy of AI, but like
• 28:40
  it kind of got it right so far. Like you kind of look at it,
• 28:43
  you're like. Yeah, I don't see the AI being
• 28:45
  our replacement. It's a human talking interface.
• 28:46
  I see the AI even trying to be our replacement.
• 28:48
  No, but I could see I could. How is quite sinister?
• 28:51
  How has his own an agenda to the degree that he cares about the
• 28:55
  outcome of something? I don't think it I really will
• 28:57
  ever care about the outcome of something.
• 28:59
  It just wants to do what you tell it to do.
• 29:00
  You haven't used croc. Have you?
• 29:02
  No, I do my best to stay away from Twitter as much as I can.
• 29:05
  I mean, I just, I feel like the best analog we have in modern
• 29:09
  day for Hal 9000 is croc in spicy mode where it's just like,
• 29:11
  Nope, we're going to Jupiter and you're like, we're all gonna
• 29:13
  die. He's like, don't.
• 29:13
  Care how my problem? When we get there, he's like,
• 29:15
  I'll figure it out. I have infinite resource.
• 29:18
  I don't need you. You are just a source of heat.
• 29:19
  Open the pod door, Hal. Not gonna do.
• 29:21
  It by the way, I mean that's what I'm really saying is it's
• 29:24
  it's prophetic in that way when I ask Siri, you know, to turn on
• 29:27
  the light and it's like I'm sorry, there are no lights.
• 29:29
  I'm like I literally just set up a light.
• 29:30
  What do you what, just turn on the light and it's like, I can't
• 29:32
  do that Space Odyssey yes, I definitely want to check.
• 29:34
  I'm going to put that on the big screen in the next couple of
• 29:36
  days and. Watch it occasionally in various
• 29:37
  parts of the world. There will be screenings of
• 29:39
  this. And if you can see it on the
• 29:40
  screen, it's even more fun. Yeah.
• 29:41
  Tarantino's theater? Yeah.
• 29:42
  I've never seen it on the theater.
• 29:43
  I would love to see it in the theater.
• 29:44
  Oh, it's fantastic. Crazy that is the episode.
• 29:47
  Thank you for joining us for all this nonsense.
• 29:48
  It's really terrible podcast from the awful company.
• 29:50
  This is in the way, but nonsense, not productions.
• 29:52
  Hi, I'm CJ Little. I'm Jeff Parker.
• 29:54
  If you like this program, please follow, download, subscribe and
• 29:56
  like an Apple Podcast, YouTube Music, Amazon Music, Antenna
• 29:58
  Pod, iHeartRadio, Spotify, My favorite Overcast where we may
• 30:01
  get your podcast from. Nonsense Studios are automated
• 30:04
  by Home Assistant, Open source home automation that puts local
• 30:06
  control and privacy 1st and is powered by a worldwide community
• 30:09
  of tinkerers and DIY enthusiasts.
• 30:11
  Forget Alexa, Home Kit, SmartThings and Google
• 30:13
  Assistant. Those are children's toys
• 30:14
  compared to Home Assistant. We'll be here every Tuesday and
• 30:16
  Thursday morning. For more nonsense, please join
• 30:18
  us. My biggest frustration with AI
• 30:32
  right now is that it's just not coming along fast enough.
• 30:34
  You can see where the future is going.
• 30:36
  I just love that you say that and.
• 30:38
  Oh man, I cannot wait to get farther forward.
• 30:40
  Yeah, my biggest frustration is I don't have enough memory to
• 30:42
  store the larger LLM model in my Fleshlight.
• 30:48
  And that's when they shut the podcast down.