Latest / How I Work / What is tokenomics, and why should you care
Transcript
- 0:00Speaker 1: Your mid task.
- 0:02Speaker 2: You've been using AI all morning and then it happens, Sorry,
- 0:09Speaker 2: you've reached your limit. Come back later is one of
- 0:12Speaker 2: the more frustrating experiences in an otherwise pretty productive workflow,
- 0:18Speaker 2: and if you're an organization that is paying for enterprise
- 0:21Speaker 2: AI licenses, that limit isn't just an inconvenience, it could
- 0:26Speaker 2: be a very expensive surprise at the end of the month.
- 0:30Speaker 2: In this episode, Neo and I get into the world
- 0:33Speaker 2: of tokenomics, which is basically the economics of how you
- 0:37Speaker 2: use your AI. We cover how usage limits actually work
- 0:41Speaker 2: across the four main platforms, and in particular, how to
- 0:45Speaker 2: time your claud sessions so you don't burn through your
- 0:48Speaker 2: window during your busiest hours. We also chat about what
- 0:52Speaker 2: enterprise teams need to know before their IT budget disappears
- 0:58Speaker 2: in just two months into the calendar. By the end
- 1:02Speaker 2: of this episode, you will have a much clearer picture
- 1:05Speaker 2: of how to get more out of your AI without
- 1:08Speaker 2: constantly hitting walls or racking up a bill that you
- 1:12Speaker 2: didn't see coming. Welcome to how IAI with me Doctor
- 1:22Speaker 2: Amantha Imba and Neo Applin, head of Inventium AI. Each
- 1:27Speaker 2: episode we share one practical way to use AI better
- 1:30Speaker 2: at work and in life. No fluff, no tech jargon,
- 1:34Speaker 2: just things you can use straight away.
- 1:37Speaker 3: Neo.
- 1:38Speaker 2: There is a term that I am finding more and
- 1:41Speaker 2: more people are using in AI circles, and that term
- 1:45Speaker 2: is tokenomics. Can you share what on earth this means?
- 1:51Speaker 3: Tokens economics, the economics of the use of your tokens.
- 1:56Speaker 1: So this is about how.
- 1:56Speaker 3: Much you can use your AI in a particular time frame,
- 2:00Speaker 3: or how many uses you get per day, per window,
- 2:02Speaker 3: those kind of things, because AI comes at a cost
- 2:06Speaker 3: to you as a usage window or how much you
- 2:09Speaker 3: can actually use in a particular time but also to
- 2:11Speaker 3: the companies themselves, and of course they want to limit
- 2:14Speaker 3: how much you can use AI so you can't hammer
- 2:16Speaker 3: it forever without them without.
- 2:18Speaker 1: Them going broke effectively.
- 2:20Speaker 3: So it represents a little differently in each of the tools.
- 2:24Speaker 3: So in chatcha, BT and copilot, generally it'll just tell
- 2:27Speaker 3: you sorry, you've used it enough for this thing for
- 2:30Speaker 3: a while, come back in a couple of hours, and
- 2:32Speaker 3: they don't really tell you how many uses you've got.
- 2:35Speaker 3: But basically the thinking models are the ones that cost
- 2:37Speaker 3: a little bit more, and they'll tell you after a
- 2:39Speaker 3: while you've been using it too much. But Claude really
- 2:42Speaker 3: has changed this discussion because they show their usage windows.
- 2:46Speaker 3: So if you actually go to Claude, you can go
- 2:48Speaker 3: to your settings and you can actually see how much
- 2:51Speaker 3: usage you've got in a session, which is a five
- 2:55Speaker 3: hour window, and also in a weekly window as well,
- 2:59Speaker 3: and so there's a bit of a italy show you
- 3:00Speaker 3: how much I've.
- 3:01Speaker 1: Used and things like that.
- 3:02Speaker 3: So if you run out of uses, in other words,
- 3:04Speaker 3: you've used up that bar, it's going to say sorry,
- 3:07Speaker 3: come back later. And this is for a regular plan,
- 3:10Speaker 3: but if you're on an enterprise plan, they're going to
- 3:12Speaker 3: be charging you for every one of your uses. That's
- 3:14Speaker 3: why this teconomics things are really coming to play. It's
- 3:16Speaker 3: about the money. What is it in chat cheptopt has
- 3:23Speaker 3: they've also got they've got units and credits. Believe it
- 3:26Speaker 3: or not, there's actually a page here. But what they've
- 3:29Speaker 3: got is for their new models that you can use
- 3:32Speaker 3: up that at a faster rate. But they don't come
- 3:34Speaker 3: out and tell you I'm terribly sorry you've used it
- 3:37Speaker 3: too much in that window.
- 3:38Speaker 1: On a bar. They actually just say sorry, there's too much.
- 3:41Speaker 3: It's a little bit of chuice to see that in
- 3:43Speaker 3: chat JEPT, but I wouldn't be surprised if in the
- 3:45Speaker 3: future they're going to come up and actually show you
- 3:47Speaker 3: on a bar how much usage you've got left. But
- 3:49Speaker 3: basically it comes down to this, if you're using the
- 3:52Speaker 3: smartest model, So if you using the thinking models in
- 3:55Speaker 3: chetipt and copilot, and if you're using say oh in Claude,
- 4:01Speaker 3: then you will use that usage more so it'll burn
- 4:06Speaker 3: your tokens.
- 4:06Speaker 1: At a faster rate.
- 4:08Speaker 3: The other thing is if you're having really long conversations,
- 4:11Speaker 3: it's going to burn that faster as well. So big
- 4:14Speaker 3: models with long conversations are going to burn that limit faster.
- 4:17Speaker 3: So only use the big models when you need to
- 4:20Speaker 3: and also try to have smaller conversations is a bit
- 4:22Speaker 3: of a tip there.
- 4:24Speaker 2: Okay. So I feel like there's two levels here, because
- 4:26Speaker 2: there's individuals that are on you know, the twenty or
- 4:30Speaker 2: thirty dollars monthly plan of any of the main AI platforms,
- 4:34Speaker 2: and then there's also a lot of listeners that are
- 4:36Speaker 2: working in organizations and they're on the enterprise version of
- 4:39Speaker 2: the plan, and that's a little bit different. So firstly,
- 4:42Speaker 2: for individuals who've just got their own personal plan that
- 4:46Speaker 2: they're paying for What are three or four things that
- 4:50Speaker 2: they can do to avoid hitting their token limits every
- 4:56Speaker 2: five hours or day every week.
- 4:59Speaker 3: Number one, check the model you're using. So if you're
- 5:02Speaker 3: doing small, light fast things, don't get the heavy duty
- 5:05Speaker 3: model to be able to help you with that. If
- 5:07Speaker 3: you need a bicycle, use the bicycle, don't go and
- 5:10Speaker 3: get the tank for that. So summarize this sentence. You
- 5:13Speaker 3: don't need opus for that. You don't need a thinking.
- 5:15Speaker 1: Model for that.
- 5:16Speaker 3: It'll actually take longer to give your response anyway, gets
- 5:18Speaker 3: quite annoying after while. Pick the right model is the
- 5:21Speaker 3: first tip. The second is, as I mentioned before, don't
- 5:24Speaker 3: have infinitely long chat threads, particularly with the big models.
- 5:27Speaker 3: They'll get dumber over time because I'll forget things. There's
- 5:30Speaker 3: so many things in its brain that needs to keep
- 5:32Speaker 3: in track of for that chat thread. But also it
- 5:35Speaker 3: costs you each time what it actually does that grabs
- 5:37Speaker 3: the entire conversation and sends that to AI pretty much
- 5:41Speaker 3: for every single chat you have. There's little bit of
- 5:45Speaker 3: exceptions in there, but generally that's a nice simplification there.
- 5:48Speaker 3: The other thing is do check, particularly if you're on CLAUD,
- 5:51Speaker 3: check that usage window to actually see how much usage
- 5:55Speaker 3: you've got, So if you're hitting the end of it,
- 5:57Speaker 3: then obviously you might want to change your behaviors to
- 6:01Speaker 3: not do as much or plan how much you're doing
- 6:03Speaker 3: in that session. And the last is plan your sessions.
- 6:07Speaker 3: So one of the things I do is as soon
- 6:11Speaker 3: as I get to my desk in the morning, I
- 6:13Speaker 3: will send a quick message or a quick thing to Claude,
- 6:16Speaker 3: which is not a problem because I'm often using Claude
- 6:18Speaker 3: first thing of the day anyway, but a lot of
- 6:20Speaker 3: people are even making jokes. I wake up in the morning,
- 6:23Speaker 3: I don't reach and read the news on my phone.
- 6:26Speaker 3: I don't scroll the Internet. Instead of what I do
- 6:29Speaker 3: is I talk to Claude. And the reason for that
- 6:31Speaker 3: is if you have the first chat with Claude, then
- 6:34Speaker 3: that starts your five hour window. So you could start
- 6:37Speaker 3: that five hour window so then it ends around about
- 6:39Speaker 3: your lunchtime, which is great, so you've got like a
- 6:41Speaker 3: morning session and an afternoon session. Whereas if you've had
- 6:43Speaker 3: the first chat with Claude at I don't know, ten
- 6:45Speaker 3: in the morning, guess what, there's your five hour window
- 6:47Speaker 3: until two pm. And maybe that's your busiest part of
- 6:50Speaker 3: the day and you kind of use it all up.
- 6:51Speaker 3: So watch those windows is another thing. So a bit
- 6:54Speaker 3: of a tip send it he little hello, welcome morning message.
- 6:58Speaker 3: Or another one is you might get a schedule in
- 7:01Speaker 3: Claude if you've got clawed cowork where it might give
- 7:03Speaker 3: you a daily report, get that happening at the start
- 7:06Speaker 3: of the day and I kick that off. Obviously, don't
- 7:08Speaker 3: burn all your tokens on that very first start of
- 7:10Speaker 3: the morning report. But that's one of the things that
- 7:12Speaker 3: you can do to be able to start those limits,
- 7:14Speaker 3: but also really watch those limits. Go to your settings,
- 7:17Speaker 3: go to the usage watch them.
- 7:19Speaker 1: Okay.
- 7:19Speaker 2: So that's claud Is there an equivalent trick if you're
- 7:23Speaker 2: in co pilot, Gemini or chat topt.
- 7:28Speaker 3: Less so because they don't have a five hour window.
- 7:32Speaker 3: It's more about only use the pro models. So I'm
- 7:36Speaker 3: using GPT. They've got instant thinking and pro, so use
- 7:40Speaker 3: the pro models when you're doing something important strategy review
- 7:43Speaker 3: your stuff, really SmackDown your work great, you really want
- 7:45Speaker 3: pro to be able to do that, But everyday stuff
- 7:47Speaker 3: you're probably in thinking. The thing I would suggest for
- 7:53Speaker 3: everyone is check your defaults. So often what happens is
- 7:57Speaker 3: you pick the model from that model pick list and
- 8:00Speaker 3: then we leave it there kind of becomes out of side,
- 8:02Speaker 3: out of mind. And so if you've picked one of
- 8:04Speaker 3: those pro models, you're then all day and that a
- 8:07Speaker 3: burns three of your usage.
- 8:09Speaker 1: So I might say come back in a couple of hours.
- 8:10Speaker 3: You've used me too much, or it'll just take too
- 8:13Speaker 3: long to give you those responses back, and then people
- 8:15Speaker 3: get frustrated and you're actually getting all this use for nothing.
- 8:19Speaker 3: So yeah, check your defaults as well. And all of these,
- 8:23Speaker 3: by the way, these are the suggestions that as you suggested,
- 8:25Speaker 3: for just the normal knowledge worker who's on like their
- 8:28Speaker 3: thirty bucks a month plan very much for those people
- 8:32Speaker 3: enterprise plan people, the suggestions are slightly different.
- 8:35Speaker 2: Okay, let's talk about enterprise. Can you explain how the
- 8:39Speaker 2: billing is different for enterprise plans because it is very
- 8:43Speaker 2: different from your own personal plan.
- 8:47Speaker 3: Yeah, and it depends on the tool as well, and
- 8:49Speaker 3: what your company has bought, and also how big your
- 8:52Speaker 3: company is is another thing as well. So back in
- 8:55Speaker 3: the day, in the dim dark day ages of like
- 8:57Speaker 3: a year ago, you would buy thirty dollars a month plan,
- 9:01Speaker 3: and if you're an enterprise you would actually just still
- 9:03Speaker 3: be paying thirty dollars a month for each of your
- 9:06Speaker 3: three thousand staff members, just say you're a big company,
- 9:09Speaker 3: and then they can use it as much as they want.
- 9:11Speaker 3: And of course if they use it a bit too much,
- 9:13Speaker 3: the tool might go tot saws come back in an
- 9:15Speaker 3: hour or two and not a problem. That's all kind
- 9:17Speaker 3: of cool. The plans are now changing, particularly led by
- 9:21Speaker 3: Claude here, and what they're doing is they're doing a
- 9:24Speaker 3: per use plan. So some companies are still on the
- 9:27Speaker 3: old way of working, still thirty bucks a month.
- 9:29Speaker 1: If you use it too much it goes tot saws
- 9:31Speaker 1: come back.
- 9:32Speaker 3: But the newer plans are you're paying a certain dollars
- 9:35Speaker 3: per month, but that doesn't include your usage, not always anyway.
- 9:40Speaker 3: And so if you're on this usage plan, what this
- 9:43Speaker 3: means is every chat thread you use will then go
- 9:46Speaker 3: to that token usage.
- 9:48Speaker 1: It's like a page to play.
- 9:49Speaker 3: So it's like a cent here, a couple of cents,
- 9:51Speaker 3: they're a buck there, those kind of things, and these
- 9:54Speaker 3: things add up.
- 9:55Speaker 1: They really do add up.
- 9:57Speaker 3: Now with the token normis I'll just say herempanies want that.
- 10:01Speaker 3: There are big tech companies here. They've even got a
- 10:03Speaker 3: leaderboard of who's using more tokens because they reckon their
- 10:06Speaker 3: programmers using AI is a good thing. I don't think
- 10:08Speaker 3: that's a great way to go. I think it really
- 10:10Speaker 3: should be use AI so it's effective for your use.
- 10:15Speaker 3: And also watch out on those billings. If you're in
- 10:17Speaker 3: an enterprise and you're on this per use plan, you
- 10:20Speaker 3: can burn through an awful lot of cash really fast.
- 10:24Speaker 3: And there are certainly in the forums I'm in part
- 10:27Speaker 3: of some companies getting massive bills that they didn't expect,
- 10:32Speaker 3: burning through their IT budget in the first two months
- 10:34Speaker 3: of the year just because of their AI token use.
- 10:39Speaker 3: So if you happen to be on one of these
- 10:41Speaker 3: company plans where it's a per use thing, if you're
- 10:44Speaker 3: the company manager, definitely review that whoever's responsible to it,
- 10:49Speaker 3: get them to do reviewing, particularly per week at the moment.
- 10:52Speaker 3: You may even want to do it per day if
- 10:53Speaker 3: it's new for your staff members so they're not going
- 10:55Speaker 3: crazy on these systems. You may want to do a
- 11:00Speaker 3: portion of costs of the departments so that each department
- 11:03Speaker 3: then is paying for what they're using. But also then
- 11:06Speaker 3: you need to train people on decent token usage mechanisms,
- 11:09Speaker 3: so have smaller chat threads to use the right model
- 11:11Speaker 3: and all those kind of things. Otherwise they're just going
- 11:13Speaker 3: to burn the tokens and not know that they're doing
- 11:15Speaker 3: that as well.
- 11:16Speaker 1: So it's like all.
- 11:17Speaker 3: Across the organization, everyone needs to watch out for their
- 11:20Speaker 3: tokens if it's a per use a system.
- 11:24Speaker 2: And I think for me, this really speaks to the
- 11:26Speaker 2: importance of training. Do not give people AI licenses without
- 11:32Speaker 2: the training, because it is going to get very very
- 11:36Speaker 2: expensive if people are not understanding tokenomics and not using
- 11:41Speaker 2: their AI in a smart way. So Neo, thank you
- 11:44Speaker 2: so much for explaining what is a term that is
- 11:48Speaker 2: coming up more and more I'm finding and discussions I'm
- 11:51Speaker 2: in around AI, and hopefully this might get people being
- 11:55Speaker 2: a little bit smarter with their token usage. How aisted
- 12:00Speaker 2: by me, Amantha Imber and Neo Applan. A big thank
- 12:03Speaker 2: you to Martin Imber who does our sound editing, and
- 12:06Speaker 2: Jim Rubio for production support, and thank you to John
- 12:09Speaker 2: Kilby who composed the theme music.