Latest / How I Work / The three agents every knowledge worker should have
Transcript
- 0:00Speaker 1: Hey, just a quick one from me before we get
- 0:02Speaker 1: into it. Imagine finishing the week with hours back because
- 0:06Speaker 1: the repetitive stuff is being handled by agents that you built.
- 0:10Speaker 1: That's what happens at the AI Agent boot Camp. It's
- 0:12Speaker 1: a four week virtual program that Neo and I built
- 0:15Speaker 1: for people already using AI who know they could be
- 0:18Speaker 1: getting more out of it. Spots are limited and details
- 0:21Speaker 1: are in the show notes. You know that feeling when
- 0:24Speaker 1: you need to research something for work, like a new market,
- 0:27Speaker 1: a regulatory change, a software purchase, and you find yourself
- 0:31Speaker 1: either drowning and browser tabs or getting back a generic
- 0:34Speaker 1: AI response that doesn't quite hit the mark. And then
- 0:37Speaker 1: there's the other end of the spectrum. You've written something
- 0:40Speaker 1: important you know probably needs a second set of eyes,
- 0:43Speaker 1: but you send it anyway because getting proper feedback feels
- 0:46Speaker 1: like it would take more time than it's worth. In
- 0:49Speaker 1: today's episode, Neo and I are covering three agents we
- 0:53Speaker 1: think that every knowledge worker really really needs that cover
- 0:57Speaker 1: these problems and more, and two of these agents we
- 1:01Speaker 1: are giving away for free. You can find links to
- 1:04Speaker 1: those in the show notes, and by the end of
- 1:07Speaker 1: this episode you'll have tools that took Neo days to build,
- 1:11Speaker 1: ready to use immediately. Welcome to how IAI with me
- 1:21Speaker 1: Doctor Amantha Imber and Neo Applin, head of Inventium AI.
- 1:26Speaker 1: Each episode we share one practical way to use AI
- 1:29Speaker 1: better at work and in life. No fluff, no tech jargon,
- 1:34Speaker 1: just things you can use straight away. If you're a
- 1:38Speaker 1: knowledge worker, there are some must have agents that Neo
- 1:44Speaker 1: and I think, Look, every knowledge worker really really needs these.
- 1:49Speaker 1: So we're going to start with the research agent. So, Neo,
- 1:54Speaker 1: what is a research agent? Why is this so important
- 1:58Speaker 1: for knowledge workers?
- 2:00Speaker 2: Researching is important for all of us to keep up
- 2:02Speaker 2: to date with the things that we do. So sometimes
- 2:05Speaker 2: it's keeping up today with the industry in news. Sometimes
- 2:07Speaker 2: it's regulations you need to keep up to date with.
- 2:10Speaker 2: But also it's about solving the problem. Maybe are buying
- 2:12Speaker 2: some software. What's good software? Maybe going into a new market,
- 2:16Speaker 2: and so what's that market or there's product about we
- 2:19Speaker 2: Google is huge for a reason. It's so you can
- 2:21Speaker 2: learn things, and AI is amazing to be able to
- 2:23Speaker 2: help you learn things. But if you ask a pretty
- 2:26Speaker 2: average query, you're going to get a pretty average response
- 2:30Speaker 2: back out of AI. It doesn't kind of read between
- 2:33Speaker 2: the lines brilliantly when it comes to research and so like.
- 2:37Speaker 2: For example, one of the things I care about is
- 2:39Speaker 2: when I do research using AI, I wanted to ownly
- 2:41Speaker 2: search sometimes reputable sources like government sources, industry sources, not
- 2:46Speaker 2: neo's blog of some rando with some opinion. I want
- 2:50Speaker 2: to get the really good stuff out of it. And
- 2:52Speaker 2: so if I don't tell the AI that I only
- 2:56Speaker 2: care about these kind of sources, then it can search anywhere.
- 3:00Speaker 2: So how do we brief our research assistant? And I
- 3:03Speaker 2: want you to think about AI as a research assistant
- 3:05Speaker 2: rather than you do the work. No, no, no, you're an
- 3:07Speaker 2: assist in doing the work. So we're going to brief
- 3:09Speaker 2: them correctly so they can do a great job of research.
- 3:12Speaker 1: Why do we need an agent to do that? I mean,
- 3:14Speaker 1: can't we just write a really good prompt into AI.
- 3:17Speaker 2: Yeah, absolutely you can. It's about how much time it
- 3:20Speaker 2: takes do you write these prompts? So you could just
- 3:23Speaker 2: say I want to know the latest on Let's go
- 3:28Speaker 2: for something a bit more legal and regulatory type thing
- 3:31Speaker 2: that some people may want to employment law. Here we
- 3:34Speaker 2: go employment law for this state. Maybe that's one of
- 3:36Speaker 2: the things you may want to go to, or maybe
- 3:39Speaker 2: there's a new product or service out there you want
- 3:41Speaker 2: to have a look at. So you can totally do
- 3:42Speaker 2: that and just write a pretty prompt. But here's the thing.
- 3:46Speaker 2: First off, most people write their search prompts like they
- 3:48Speaker 2: search with Google, so they don't give enough background and
- 3:52Speaker 2: enough context. They don't actually say where they would like
- 3:55Speaker 2: to search, what the real goal of this search thing
- 3:57Speaker 2: is about. And so you'll get outputs. But is it
- 4:01Speaker 2: a great output or is it kind of like you've
- 4:02Speaker 2: just gone to Google you're the ten top ten results
- 4:05Speaker 2: and summarize them. Likely the latter.
- 4:08Speaker 1: So I know that you have a research agent, and
- 4:12Speaker 1: for the very fortunate how I work listeners, we're actually
- 4:17Speaker 1: going to be giving that away. So there is a
- 4:20Speaker 1: link to get Neo's research agent in the show notes, Neo,
- 4:24Speaker 1: what are the ingredients that you've put into your research agent?
- 4:28Speaker 1: Because I know you've put a lot of work into
- 4:30Speaker 1: this one.
- 4:31Speaker 2: Its goal is to be that assistant, that helper to
- 4:35Speaker 2: figure out what you're really trying to get to and
- 4:38Speaker 2: then to build you out an effective prompt. So you
- 4:41Speaker 2: start off with a basic prompt, here's what I wanted
- 4:43Speaker 2: to research. What I want to research? A bit of
- 4:44Speaker 2: a goal or a background and all those kinds of things.
- 4:47Speaker 2: It will then flesh that out into a way that
- 4:49Speaker 2: you can get a great result. So it might say,
- 4:52Speaker 2: as I mentioned before, the sources that you might have
- 4:54Speaker 2: looked at. If it gets the vibe that you're after
- 4:57Speaker 2: something like you know, laws, regulations, then it will go
- 5:00Speaker 2: for the serious stuff. And I'm talking about like government sources.
- 5:04Speaker 2: Maybe it might be tier one consulting firms are like
- 5:06Speaker 2: if you're going for legal stuff, that's a very different thing.
- 5:09Speaker 2: It'll go for very much legal things. If it's about
- 5:11Speaker 2: product things, then it will probably go to what's those
- 5:14Speaker 2: products in those markets and then be able to do
- 5:15Speaker 2: the research for you. So it's more about understanding what
- 5:18Speaker 2: you're trying to achieve with your search, and what it
- 5:21Speaker 2: will do is then I'll put more guardrails, more binds
- 5:23Speaker 2: on it, and also tell the AI the types of
- 5:26Speaker 2: things that it needs to research and then the types
- 5:29Speaker 2: of things that it needs to have in the final report.
- 5:32Speaker 2: These are also quite important things because there's no point
- 5:35Speaker 2: reading how bunch of pages if it's in a way
- 5:37Speaker 2: it's not going to help you. So it will suggest
- 5:39Speaker 2: these kind of things. Now it produces a usually a
- 5:43Speaker 2: little bit longer of a prompt than the one you've
- 5:45Speaker 2: put in. It's certainly a lot more specific. With these
- 5:47Speaker 2: kind of rules, you can tweak that, you can do
- 5:49Speaker 2: whatever you need to it, and then you can get
- 5:50Speaker 2: that and put that into your AI to be able
- 5:53Speaker 2: to do a search. Now, this I've found is an
- 5:56Speaker 2: awesome way to use it, even if you're using one
- 5:58Speaker 2: of those deep research tools. So it's like researcher and
- 6:01Speaker 2: deep research once again, if you give them a great
- 6:04Speaker 2: prompt into that, you will even get a better response out.
- 6:07Speaker 2: But the prompt in makes such a difference.
- 6:11Speaker 1: Let's move into the critical thinking agent, which again is
- 6:17Speaker 1: one that well, I think we all use that a
- 6:19Speaker 1: lot at Inventium all the time. Why do knowledge workers
- 6:24Speaker 1: need a really good critical thinking agent.
- 6:27Speaker 2: It's like a bit of a check before you give
- 6:29Speaker 2: anything to anyone. Isn't it great to have a bit
- 6:31Speaker 2: of a check over it? Because I don't know about you,
- 6:33Speaker 2: but maybe you need a second cup of coffee in
- 6:35Speaker 2: the morning, and sometimes your brain takes a little bit
- 6:37Speaker 2: to get going. Other times we just forget stuff that
- 6:41Speaker 2: really should be there, or other times, if you're like
- 6:44Speaker 2: me and you've got your brain going with like forty
- 6:46Speaker 2: seven tabs opening your brain when you're working on something,
- 6:49Speaker 2: maybe you just forgot something or you didn't quite explain
- 6:51Speaker 2: it particularly well, or as a bad cut and paste.
- 6:53Speaker 2: It's like a pre check before you give anything to
- 6:56Speaker 2: anyone else. So kind of like I make the joke
- 6:59Speaker 2: that we don't give anything to anyone without doing a
- 7:01Speaker 2: spell check first. Why don't we get it to do
- 7:03Speaker 2: like a quick critical check before we give it to
- 7:05Speaker 2: anyone else. That way, we're not wasting our time or
- 7:08Speaker 2: their time, and all the work that we do that
- 7:10Speaker 2: we do at least comes off to being really good.
- 7:12Speaker 2: So it's an AI check of our work before we
- 7:15Speaker 2: pass it to anyone else.
- 7:17Speaker 1: So I know that, you know a lot of people
- 7:20Speaker 1: will recommend, including us, that when we're working with AI,
- 7:24Speaker 1: to do something that is at least medium to high stakes,
- 7:26Speaker 1: Like we're not talking, just write an email response to Bob,
- 7:30Speaker 1: Like you know, it might be a report that you
- 7:31Speaker 1: want checked. Why can't you just go, oh, you know, AI,
- 7:37Speaker 1: can you review this and point out any problems with
- 7:41Speaker 1: my argument and ways that I can improve this? Like
- 7:43Speaker 1: is there really such a big difference between writing a
- 7:46Speaker 1: really basic prompt like that, and having this critical thinking
- 7:50Speaker 1: agent critique our work.
- 7:52Speaker 2: Yes and no, it's up to you how deep you
- 7:54Speaker 2: want any kind of critique to go. Really like, if
- 7:57Speaker 2: it's a small pace of working, is it any good?
- 7:59Speaker 2: Then you'll probably get an okay response back from AI.
- 8:02Speaker 2: But sometimes it's easier to have a well thought out
- 8:05Speaker 2: prompt will actually go through it from many different dimensions.
- 8:07Speaker 2: So it's not just does it read well? Is it
- 8:11Speaker 2: logically sound? Like? If I made an argument in there,
- 8:13Speaker 2: does it actually make sense? Have I made up facts
- 8:17Speaker 2: which just don't make sense? I'd want AI to say,
- 8:20Speaker 2: you know, I think that's a bit wrong. So what
- 8:23Speaker 2: we've got is this critical thinking buddy if you'd like
- 8:26Speaker 2: to be able to critique your stuff, but it looks
- 8:28Speaker 2: at it from many different dimensions there, and it'll also
- 8:30Speaker 2: tell you where there might be some problems and suggestions
- 8:34Speaker 2: on how you can fix it. Now, you can absolutely
- 8:36Speaker 2: rite all that yourself, but that's a lot of typing
- 8:38Speaker 2: every single time you want to do a prompt for
- 8:40Speaker 2: it to review the stuff. So it's just easy to say, look,
- 8:43Speaker 2: I'll use my all purpose checker to be able to
- 8:46Speaker 2: check the whole lot. Now I may not agree with everything.
- 8:48Speaker 2: I might say yeah, I'm fine with that, and that's okay,
- 8:51Speaker 2: but you might get one or two points out of it.
- 8:53Speaker 2: You go, wow, that's amazing, And so you can fix
- 8:55Speaker 2: these things before you send it out and so much
- 8:57Speaker 2: faster for you to get a quick check because you've
- 8:58Speaker 2: already got a pre bake as your agent.
- 9:02Speaker 1: I do use this agent very frequently and for how
- 9:05Speaker 1: I work listeners. Again, there's a link in the show notes.
- 9:08Speaker 1: We are giving this one away for you, and it
- 9:12Speaker 1: is essentially like what you'll get with these with these
- 9:16Speaker 1: agents is a pretty long comprehensive system prompt that you
- 9:20Speaker 1: can copy and paste into your GPT if you're in CHATJPT,
- 9:24Speaker 1: in your agent, if you're in copilot your GM and
- 9:26Speaker 1: Gemini or a project in terms of your project instructions
- 9:30Speaker 1: in Claude. Now, there is one more that we certainly
- 9:35Speaker 1: believe every knowledge worker should have in terms of agents,
- 9:38Speaker 1: and that is a cross functional advisory board. Now, can
- 9:42Speaker 1: you tell us what is this one all about?
- 9:46Speaker 2: A lot of people who are in leadership positions, they've
- 9:49Speaker 2: got their own advisory board. Now, maybe you're a CEO
- 9:52Speaker 2: of an SM or a medium sized company. He you're
- 9:56Speaker 2: definitely if you're running a big company, you'll have people
- 9:58Speaker 2: you can phone to be able to say, Hey, I've
- 10:00Speaker 2: got this big problem, this big strategy thing that i'm doing.
- 10:03Speaker 2: What do you think am I getting it all right?
- 10:06Speaker 2: So it's kind of the stakes are higher than just
- 10:09Speaker 2: here's a report, quickly check it. That's what the critical
- 10:11Speaker 2: Thinking Buddy is great for. But sometimes it's like, I
- 10:14Speaker 2: really want to make sure I'm looking at this from
- 10:16Speaker 2: different perspectives. So if you're doing like a strategy stuff,
- 10:20Speaker 2: then it depends on which company you're in and what
- 10:22Speaker 2: the strategy is about, but you might need to have
- 10:24Speaker 2: someone looking at it from a people lens, maybe there's
- 10:27Speaker 2: a product lens, maybe there is an operational lens that
- 10:31Speaker 2: you need to look at these things. So it's kind
- 10:34Speaker 2: of like, wouldn't it be great if you had your
- 10:35Speaker 2: own team of experts be able to help you to
- 10:38Speaker 2: review what you've done and then say, look, from our perspective,
- 10:41Speaker 2: it looks good, or you can tweak it, or you
- 10:43Speaker 2: can improve it in different ways. So what this is
- 10:45Speaker 2: is AI being your team of experts to review stuff
- 10:49Speaker 2: from different perspectives. Really it's the stuff that matters, so
- 10:52Speaker 2: you get a much better input from them and you
- 10:55Speaker 2: can tweak your stuff if you need to.
- 10:57Speaker 1: And I know with this one, we are giving this
- 11:00Speaker 1: away to everyone that enrolls in Inventium's AI Agent boot Camp,
- 11:04Speaker 1: which kicks off in July. We've got a virtual program
- 11:08Speaker 1: a running which anyone can join from anywhere they like.
- 11:11Speaker 1: We've also got in person workshops running in Melbourne and Sydney,
- 11:15Speaker 1: and so anyone enrolling in that will get this cross
- 11:19Speaker 1: functional advisory board agent, which is awesome, you know, But
- 11:25Speaker 1: I'm curious, neo, because I know that there's a lot
- 11:27Speaker 1: of people that are like, oh, I've built lots of agents,
- 11:29Speaker 1: and you know, it only takes me like five or
- 11:31Speaker 1: ten minutes to build an agent. And yes there are
- 11:33Speaker 1: some really simple ones that can be built in a
- 11:35Speaker 1: short amount of time. But just because when I asked
- 11:39Speaker 1: you this, I was quite surprised by your response. How
- 11:41Speaker 1: long did it actually take you to build these three
- 11:44Speaker 1: agents that I know? You know, the team and Inventium
- 11:47Speaker 1: certainly use a lot. You use a hell of.
- 11:49Speaker 2: A lot each h It's hard to gauge because some
- 11:54Speaker 2: of them I've been building and then refining over over time.
- 11:57Speaker 2: Because there's a couple of challenges the first is how
- 11:59Speaker 2: do you build an agent that will do the thing
- 12:02Speaker 2: that you need to do it. That's great to have
- 12:05Speaker 2: that thing, but then how does it do it right
- 12:08Speaker 2: every single time? Because often you've got different inputs. So
- 12:11Speaker 2: it might be a report builder agent that you're getting
- 12:14Speaker 2: to build, and that's fine for this input that I
- 12:17Speaker 2: give it today, but tomorrow's input it might give me
- 12:19Speaker 2: a slightly different outcome. So it's around about getting the
- 12:23Speaker 2: agent to actually solve the problem that you're trying to solve.
- 12:26Speaker 2: So getting it to work with AI and make sure
- 12:28Speaker 2: that you've got it working there is great. The The
- 12:30Speaker 2: other thing is you've got to test it, and a
- 12:32Speaker 2: lot of people aren't testers, and so how do you
- 12:34Speaker 2: make sure it works under these different circumstances input, different inputs,
- 12:37Speaker 2: different outputs, all those kind of things, So that takes
- 12:40Speaker 2: an awful lot of time. The third bit it takes
- 12:42Speaker 2: time also is how do we make sure that it's
- 12:44Speaker 2: still working in the future because these AI tools continually update.
- 12:49Speaker 2: It goes from JGBT version four to version five to
- 12:52Speaker 2: five point three. If I put all these different versions
- 12:54Speaker 2: give you slightly different outcomes. So I'm continually refining these
- 12:58Speaker 2: agents and I encourage you to do the same too.
- 13:00Speaker 2: So if you are building an agent, that's great. What
- 13:03Speaker 2: you can do is ask AI to see whether you
- 13:06Speaker 2: can improve that prompt at all. It might give you
- 13:08Speaker 2: some good suggestions there. But also I really encourage you
- 13:12Speaker 2: to test this, particularly if for giving your agents to
- 13:14Speaker 2: other people, because they will put in different inputs in
- 13:17Speaker 2: different words, in different ways, and they might get weird
- 13:19Speaker 2: outputs from it and go that's ageb's crap. So therefore
- 13:21Speaker 2: the testing pot's so critical. So yeah, it can take
- 13:24Speaker 2: me quite happily days for some of them that I
- 13:27Speaker 2: have built. Others is relatively faster in an hour or two.
- 13:30Speaker 2: But yeah, the important ones, like the research critical thinking
- 13:34Speaker 2: at least half a day to a date.
- 13:35Speaker 1: So there you go, and you as a listener can
- 13:39Speaker 1: get that without needing to put in half a day
- 13:42Speaker 1: or a day's worth of work. So, as I said,
- 13:44Speaker 1: go into the show notes. There is a link to
- 13:47Speaker 1: access the research agent and the critical thinking agent. And
- 13:50Speaker 1: of course if you want to come along to our
- 13:53Speaker 1: AI Agent boot camp which Neo is leading, which is
- 13:57Speaker 1: just going to be so awesome. We've had so many
- 14:00Speaker 1: poul for asking for this, then you'll also get the
- 14:02Speaker 1: cross functional Advisory Board agent in that one, so there
- 14:05Speaker 1: is a link to enroll. In the show notes. How
- 14:09Speaker 1: i AI was hosted by me Amantha Imber and Neo Applan.
- 14:13Speaker 1: A big thank you to Martin Imber who does our
- 14:15Speaker 1: sound editing, and Jim Rubio for production support, and thank
- 14:19Speaker 1: you to John Kilby who composed the theme music.