Latest / How I Work / Agents, Projects, and Skills (oh my!): A no-fluff guide on when to use each one
Transcript
- 0:00Speaker 1: If you've ever sat there wondering what the difference is
- 0:03Speaker 1: between an agent, a project, and a skill, and whether
- 0:07Speaker 1: you're supposed to be using a GPT, JEM, a notebook
- 0:09Speaker 1: or something else entirely, you are not alone. The AI
- 0:13Speaker 1: industry has done a spectacularly bad job of naming any
- 0:18Speaker 1: of this, and it makes getting started feel harder than
- 0:21Speaker 1: it should. In this episode, Neo and I try to
- 0:24Speaker 1: sort it out properly. You will learn what an agent
- 0:28Speaker 1: actually is and when to reach for one, why projects
- 0:32Speaker 1: shine when you've got a pile of documents you want
- 0:34Speaker 1: to ask questions about, and how skills work like a
- 0:38Speaker 1: recipe your whole system can follow. By the end of
- 0:42Speaker 1: this episode, you will know exactly which of the three
- 0:45Speaker 1: to use and when, with no more guessing involved. Welcome
- 0:56Speaker 1: to how IAI with me, Doctor Amantha Imba and Neo Applin,
- 1:01Speaker 1: head of Inventium AI. Each episode we share one practical
- 1:05Speaker 1: way to use AI better at work and in life.
- 1:08Speaker 1: No fluff, no tech jargon, just things you can use
- 1:12Speaker 1: straight away, Neo. As we often complain about on how
- 1:16Speaker 1: I AI, there are so many different words and terms
- 1:21Speaker 1: for things, and for someone that is not deep in AI,
- 1:24Speaker 1: every day like us, it can be really confusing, and
- 1:28Speaker 1: so today I want to unpack water skills versus water
- 1:32Speaker 1: projects versus what are agents slash GPTs slash gems. So
- 1:38Speaker 1: let's start with agents, which in Copilot we're talking about agents,
- 1:45Speaker 1: in Chat gpt we're talking about GPTs, and in Gemini
- 1:47Speaker 1: we are talking about gems. What are these things neo
- 1:51Speaker 1: and when should we use them?
- 1:53Speaker 2: Apologies on behalf of the whole AI industry. They're horrendous
- 1:57Speaker 2: at the naming on this one. So we are talking
- 1:59Speaker 2: about this. It's ish kind of thing. And really the
- 2:03Speaker 2: Copilot agent Chachapite, GPTs and Gemini gems I might just
- 2:07Speaker 2: call them in air quotes agents. These ones are ais
- 2:10Speaker 2: with a job. So you get them to do something.
- 2:14Speaker 2: You give them a task, which is generally some instructions.
- 2:16Speaker 2: You might give them a couple of templates or look
- 2:18Speaker 2: up files or something like that to be able to
- 2:19Speaker 2: answer your questions or to help do the task, and
- 2:22Speaker 2: they do that one job. There's a big asterisk here
- 2:26Speaker 2: in that it doesn't take forever to do the job.
- 2:28Speaker 2: It will not work for twenty four hours, it will
- 2:30Speaker 2: not wake up at three in the morning. You give
- 2:33Speaker 2: it something to do, it will do that thing and
- 2:36Speaker 2: we'll finish it within a minute or two generally, and
- 2:39Speaker 2: when it produces for you is generally up to about
- 2:41Speaker 2: like five pages worth of content. So if you've got
- 2:45Speaker 2: a spreadsheet with a thousand rows, it is not the
- 2:47Speaker 2: task for you. But if you say, hey, I've got
- 2:50Speaker 2: a meeting transcript that I want you to clean up
- 2:52Speaker 2: using this template, it is absolutely the job for you
- 2:55Speaker 2: to use. So that is an agent, copilot, agent, chat TOPTGBT, Gemini, GM,
- 3:01Speaker 2: an AI with a job.
- 3:02Speaker 1: Okay, so, Neo, what are maybe two or three of
- 3:07Speaker 1: the most common or the agents that you used most
- 3:12Speaker 1: often in your work life?
- 3:15Speaker 2: Wow? Okay, I am not usual because I'm always creating
- 3:20Speaker 2: agents and things like that. Where I think most people
- 3:23Speaker 2: will probably use them would be within their job. So
- 3:26Speaker 2: what it comes down to is you've got to figure
- 3:28Speaker 2: out what you're doing to figure out whether you need
- 3:30Speaker 2: an agent to help you with that. So I have
- 3:31Speaker 2: a researcher agent that helps me to specify what I'm
- 3:36Speaker 2: researching and ask me questions about that, so I can
- 3:39Speaker 2: then give that to a deep research tool. So I
- 3:42Speaker 2: get a really good research agent. Why because I need
- 3:44Speaker 2: to keep up to date with things, and I'm always
- 3:46Speaker 2: learning things, and so researching is very much part of
- 3:48Speaker 2: my job. A lot of people would probably have meeting
- 3:52Speaker 2: minute cleanup agents or gipitis and gems, and what that'll
- 3:55Speaker 2: do is you'll take maybe your meeting transcript, maybe take
- 3:57Speaker 2: your agenda, and it'll then format it into a report
- 4:00Speaker 2: or an output which might be his the items and
- 4:04Speaker 2: maybe in a table for who's doing what when and
- 4:07Speaker 2: things like that, and maybe you can even call out
- 4:09Speaker 2: these were the things that we're on the agenda that
- 4:11Speaker 2: weren't talked about. So those kind of things, meeting agents
- 4:13Speaker 2: and things like that. I've built a bunch for HR people.
- 4:17Speaker 2: So it's things like maybe reviewing job ads before you
- 4:22Speaker 2: send them out, or it might be a competency framework
- 4:25Speaker 2: and things like that which are very much detailed stuff
- 4:27Speaker 2: that you do that he needs a lot of checking
- 4:29Speaker 2: and things like that. Agent's GPT's gems are awesome for that.
- 4:34Speaker 1: Let's now talk about projects. So I know in Claude
- 4:38Speaker 1: and chat JPT they are called projects. What is the
- 4:43Speaker 1: equivalent in Gemini and co pilots.
- 4:47Speaker 2: Not an equivalent in Gemini, They've just got gems in Gemini.
- 4:51Speaker 2: In co pilot. They've kind of sort of got of equivalent. Maybe. Okay,
- 4:56Speaker 2: it's called notebooks. It's actually just gone through a bit
- 4:59Speaker 2: of a reach. Actually looks a little bit closer to
- 5:01Speaker 2: a project in the others, So I'd say, yeah, notebooks
- 5:05Speaker 2: are probably equivalent. But projects and notebooks are similar kind
- 5:08Speaker 2: of things, and they're also kind of similar to an
- 5:11Speaker 2: agent GPT and GEM in that it takes your instructions
- 5:14Speaker 2: and also particularly take a board. We'll take a lot
- 5:16Speaker 2: of your files so you can answer questions about those files.
- 5:19Speaker 2: That's how they're engineered.
- 5:21Speaker 1: So when should we use a project as distinct from
- 5:24Speaker 1: an agent?
- 5:26Speaker 2: Okay, it depends on the tool you're using. So if
- 5:29Speaker 2: you're in let's say chat GPT, because this makes it
- 5:32Speaker 2: nice and complicated because you've got GPTs, you've also got
- 5:36Speaker 2: projects and they're also got skills, which will cover in
- 5:38Speaker 2: a second. So I would use a GPT when I've
- 5:42Speaker 2: got a single task, a single job that I need
- 5:44Speaker 2: to do again and again and again. I would use
- 5:47Speaker 2: a project if I've got a lot of knowledge, a
- 5:49Speaker 2: lot of information there that I need to ask questions about,
- 5:52Speaker 2: or maybe i've got a topic like a real project
- 5:54Speaker 2: that I'm actually need to do lots of work about
- 5:57Speaker 2: because one of the great things about projects is yes,
- 5:59Speaker 2: it'll lots of documents, but also groups your chats about
- 6:03Speaker 2: those topics together. So you might have a maybe you're
- 6:08Speaker 2: implementing a new CRM, so you might have the new
- 6:10Speaker 2: CRM projects, and you'll have all the files about that
- 6:12Speaker 2: new CRM and all the chats you have about that
- 6:14Speaker 2: new CRM. Well then get grouped together, which is kind
- 6:16Speaker 2: of cool in one project. So I use projects for
- 6:20Speaker 2: things that you work on. So I've got a project
- 6:22Speaker 2: for Inventium, which has a lot of products and services
- 6:24Speaker 2: and things like that. You might have one for like
- 6:27Speaker 2: a standard operating procedures. You might have one as like
- 6:29Speaker 2: an onboarding one. You might have one standard Q and
- 6:32Speaker 2: a one up to you what you need.
- 6:34Speaker 1: So am I right in saying that with a project,
- 6:36Speaker 1: projects shine where there's heaps of context that you want
- 6:41Speaker 1: to have the AI know about you to do a
- 6:44Speaker 1: really good job in what you're asking it to do.
- 6:47Speaker 2: Absolutely. Yeah. So the more context, more backgrounds, it's very
- 6:50Speaker 2: much a document heavy, conversation heavy thing rather than a
- 6:54Speaker 2: job heavy thing. So if I'm asked getting questions as
- 6:58Speaker 2: answered from documents from knowledge. Absolutely our create a project.
- 7:02Speaker 1: Okay, finally, let's talk about skills now, just to get
- 7:06Speaker 1: the language right. In Claude, we have skills in chat GPT,
- 7:11Speaker 1: we have skills in co pilot. We also have in.
- 7:16Speaker 2: Copilot Cowork, we have skills. Believe it or not. So
- 7:20Speaker 2: this is it's come out in the last few weeks,
- 7:25Speaker 2: so you may see it on your copilot screen. So
- 7:28Speaker 2: a couple of caveats. This is only for company users.
- 7:31Speaker 2: So this is the Microsoft three six y five copilot
- 7:34Speaker 2: users and your company may or turn it on, they
- 7:37Speaker 2: may turn it off. And what it is, it's part
- 7:41Speaker 2: of the Cowork type thing. It very much feels like
- 7:45Speaker 2: claud Cowork. They've used a lot of that technology there.
- 7:48Speaker 2: And skills in copilot can only be used in copilot Cowork.
- 7:51Speaker 2: They cannot be used in regular copilot.
- 7:53Speaker 1: Great, just to make everything more confusing. And do we
- 7:56Speaker 1: have an equivalent in Gemini.
- 7:58Speaker 2: No, No skills in Gemini, although I wouldn't be surprised
- 8:01Speaker 2: if some stage it comes out. It's become an open standard.
- 8:03Speaker 2: And this is where I love the word skills, because
- 8:05Speaker 2: we're talking about the same kind of thing. In we
- 8:07Speaker 2: at least three of the four main tools. And so
- 8:10Speaker 2: skills are actually really good and they're like a reusable method,
- 8:15Speaker 2: so it's something that you can do again and again
- 8:18Speaker 2: and again. So what you will find, maybe I'll give
- 8:22Speaker 2: you an example of how these things kind of fit together.
- 8:24Speaker 2: So I want you to imagine like a restaurant kitchen.
- 8:27Speaker 2: So I love going to a restaurants where you can
- 8:29Speaker 2: actually see the kitchen and maybe you look at the
- 8:30Speaker 2: bar and watch people do their jobs, and I find
- 8:33Speaker 2: it fascinating because they're awesome. And what you find in
- 8:36Speaker 2: a restaurant kitchen is you'll have specialists in different areas.
- 8:39Speaker 2: You might have someone at the grill, they're the girl master,
- 8:42Speaker 2: and maybe you'll have the pastry chef and they're sitting
- 8:44Speaker 2: over there doing the pastry stuff. And generally what you
- 8:47Speaker 2: find is each of these cooks, these chefs just doing
- 8:51Speaker 2: one thing, if you'd like, that's pretty much like an
- 8:54Speaker 2: agent GPT in a gym. So they do one thing,
- 8:57Speaker 2: they know what they're doing, they do that thing really
- 9:00Speaker 2: really well. A skill is more like a recipe, and
- 9:04Speaker 2: the recipe is then how to do those things and
- 9:06Speaker 2: the steps you do to be able to do, you know,
- 9:09Speaker 2: making some pastry or to be able to grill a burger,
- 9:12Speaker 2: which is very different than maybe grilling some lamb chinks
- 9:15Speaker 2: or something like that. And so the skill is more
- 9:18Speaker 2: like a recipe that can be used by you or
- 9:21Speaker 2: pretty much anyone in the system.
- 9:24Speaker 1: I think one of the ways I think about skills
- 9:26Speaker 1: is like standard operating procedures. So where it's like do
- 9:30Speaker 1: this than this than this, and what I know I
- 9:34Speaker 1: think you taught me this is that where a skill
- 9:38Speaker 1: is better than a long prompt is that adherence is better. So,
- 9:44Speaker 1: particularly in some of the AI, is if I paste
- 9:46Speaker 1: in a really long prompt that say, you know, could
- 9:50Speaker 1: also be a skill. Sometimes the AI will be a
- 9:54Speaker 1: little bit hard of hearing and it might miss some
- 9:55Speaker 1: of the steps. But in a skill, am I correct
- 9:59Speaker 1: in saying that the adherence to each and every step
- 10:03Speaker 1: in that recipe or standard operating procedure will be adhered to.
- 10:09Speaker 2: With anything, there's a massive exception because you can write
- 10:11Speaker 2: good skills and bad skills and things like that. Skills
- 10:15Speaker 2: are less often less directive. They're less about do this,
- 10:18Speaker 2: this and this and this as far as the minutia
- 10:21Speaker 2: of it all, but they do have overall steps. So
- 10:24Speaker 2: you need to include these things and things like that.
- 10:27Speaker 2: Skills work really well with goals, So here's what the
- 10:29Speaker 2: outcome is supposed to be, here's some templates of what
- 10:32Speaker 2: the outcome needs to be, and things like that. Skills
- 10:35Speaker 2: often can do other things as well. So they might
- 10:38Speaker 2: be able to invoke a Python script. In other words,
- 10:40Speaker 2: it can do things with data and things like that.
- 10:43Speaker 2: It could do things with pulling together files in different
- 10:46Speaker 2: ways and things like that. So skills are very very capable,
- 10:49Speaker 2: and they can then refer to other skills as well,
- 10:52Speaker 2: so one skill can effectively hand off to another one
- 10:54Speaker 2: or recommend that the next step is the next skill,
- 10:57Speaker 2: and things like that. So yeah, skills can be very
- 11:00Speaker 2: valuable and I prefer using skills over massive prompts absolutely.
- 11:05Speaker 1: So if we were to summarize the difference between agents, projects,
- 11:10Speaker 1: and skills, how would you do that? In a nutshell,
- 11:15Speaker 1: it's more.
- 11:16Speaker 2: About what tools you've got and then how you want
- 11:19Speaker 2: to solve a work problem. So we'll start with projects first.
- 11:22Speaker 2: Projects are great at solving the problem on I've got
- 11:25Speaker 2: to get information from these documents, from these sources, and
- 11:29Speaker 2: it is excellent at doing those. And you can also
- 11:31Speaker 2: curate what goes into the documents and the sources and
- 11:34Speaker 2: things like that. So I use projects very much. For
- 11:37Speaker 2: like I've got an an inventium and personal briefing documents
- 11:40Speaker 2: about what I'm doing and what we're working on and
- 11:42Speaker 2: things like that, and they're awesome for here's the procedures
- 11:45Speaker 2: and the policies and things like that, so awesome for those.
- 11:48Speaker 2: I would use a copilt agent, CHPTGPT or a Gemini
- 11:53Speaker 2: gem to be able to do a single task. So
- 11:57Speaker 2: I've got a meeting, I need to clean it up.
- 11:59Speaker 2: I've got, As I said research before is really good
- 12:01Speaker 2: for those kinds of things. I even use it for
- 12:03Speaker 2: helping me to refine our language and prompts and things
- 12:06Speaker 2: like that. So particularly with language, I've got like a
- 12:09Speaker 2: Ventium style skill which makes things sound like our brand style.
- 12:12Speaker 2: So so not skilled agent, I should say to do that.
- 12:15Speaker 2: Most of these agents, you could also build them as
- 12:18Speaker 2: a skill as well. And so if you've got a
- 12:21Speaker 2: tool like CHPT that has GPTs and skills, the more
- 12:25Speaker 2: modern way of doing it is absolutely built as a skill.
- 12:27Speaker 2: So a skill is great. If you've got like a
- 12:30Speaker 2: step based thing, you've got some guides, you've got some
- 12:32Speaker 2: templates and things like that. The great thing about the
- 12:34Speaker 2: skill is it can be used pretty much throughout the area.
- 12:37Speaker 2: So it's almost like you've got a head chef who
- 12:39Speaker 2: can run around the kitchen and use any of the recipes,
- 12:42Speaker 2: any of the skills that they want to, as opposed
- 12:44Speaker 2: to like an agent, GPT a gem, which is just
- 12:47Speaker 2: a single cook that can only cook one thing, one recipe. Okay,
- 12:51Speaker 2: I love that.
- 12:52Speaker 1: I hope that this has been useful for people who
- 12:56Speaker 1: are just feeling a little bit more overwhelmed by all
- 12:58Speaker 1: the different to and ways to get the most out
- 13:01Speaker 1: of AI. Hopefully now you are clearer on what is
- 13:04Speaker 1: a skill, what is a project, and what is an
- 13:06Speaker 1: agent and when to use each of those three very
- 13:10Speaker 1: distinct methods in AI. How I AI was hosted by
- 13:15Speaker 1: me Amantha Imber and Neo Applan. A big thank you
- 13:18Speaker 1: to Martin Imber who does our sound editing, and Jim
- 13:21Speaker 1: Rubio for production support, and thank you to John Kilby
- 13:24Speaker 1: who composed the theme music.