Entries

An AI agent is a type of software typically driven by generative AI (and/or advanced machine learning, in some cases) that gets a specific kind of work done, with limited human oversight.

AI agents differ from older software models (e.g. SaaS) by offering more adaptive, non-tedious workflows and minimal UX – providing the minimum of user interface needed to accept text, audio, or visual inputs. They can enrich user inputs by automatically gathering and assembling useful information from other methods, such as APIs, web scrapers, integration platforms, or even other AI agents.

Similarly, an agent’s outputs may happen partially or entirely outside of a user interface traditionally associated with a software user interface.

At the core of AI agent is its ability to make a sequence of decisions in real time using generative AI, adapting its responses based on context and goals. As such, an AI agent can be modeled on a human role (e.g. a “bookkeeper agent”). By the same token, a team of AI agents (a swarm) can emulate a collaborative team of worker roles.

A note on “feedback loops”: AI agents may be designed to improve decision-making based on user interaction with their own outputs but this is a non-essential trait that may be not cost-effective to design and build.

Generative AI (Gen AI) is a type of AI that uses an counterintuitively large data sets to train itself to behave in a way that allows it to understand and emulate human data. Hence the misguided name “large language models” (LLMs). It’s misguided because Generative AI doesn’t actually speak or understand language it is exposed to; it merely recognizes statistical patterns once it translates language (or imagery, sound, etc) into data.

Gen AI is inherently incapable of autonomous thought but if properly designed, can give the impression of thought in response to human prompts.

Generative AI is a useful technology but difficult to leverage efficiently due to its the high cost of creation, training, and hosting. Also problematic is that companies which produce Generative AI products are tempted to violate copyright laws to fulfll their need to supply “training data”, appropriating content that doesn’t belong to them.

Nevertheless, Gen AI and innovations built on top of it (eg Agentic AI) are disruptive and significant.

Agentic describes a near autonomous, goal-directed software solution chiefly comprised of an AI Agent or ideally, a team thereof. Agentic AI is closely associated in practice with Generative AI, though in theory teams of AI Agents could be built with other forms of machine learning.

It is important to note that no agentic solution is actually autonomous; what’s important ultimately is its ability to accomplish tasks with minimal human supervision.

I want to expand on my latest dictionary entry – “Lead magnet”.

Caveat: as with every other definition, there’s an implication of quality. For example, when I define marketing, the scope of the definition is good marketing, not crappy marketing.

So I’m interesting in defining quality lead magnets not the crap you instantly toss in the trash – but let’s think about the latter.

A lead magnet should be very difficult to make

The standard lead magnet formula will sound familiar: it’s a cheatsheet, a 1-pager how-to, a template, a swipe list. It’s just good enough to pique curiosity.

You look at it for a few minutes and then, “meh”, you delete it. Then you get a dozen emails in automated sequence, then you’re on a list, and so on.

At some point within the last 5 years, this got very old.

The format may be fine, of course; a concise how-to manual, for example – nothing wrong with that. But the standard formula I refer to is marked by a conspicuous lack of effort.

If you created it in one day, let alone one sitting, it’s not a lead magnet. Don’t be fooled by course creators and marketing gurus who lead you down the “it’s easy path”.

A lead magnet is not for list-building

Part of the problem is a misconception as to the fundamental purpose of a lead magnet – that it’s a email list-building tool, a “list magnet”.

Maybe 10 years ago that was true, but that concept is very shaky now.

What’s the point of an email list anyway? for 99.99% of people in B2B solutions, an email list is a marketing channel that leads to sales of solutions. Granted: for a select few, the list itself is the product – Venkat, for example. But I’ve never seen such a person lure customers (ie. paid Subtack subscribers) with a traditional lead magnet.

Keep in mind, this is just my opinion. Someone could easily counter it; “of course it’s a good idea to build your list – why wouldn’t you use a lead magnet for that?”.

In my opinion, you wouldn’t because it’s a wasted opportunity to create value – more on that below but first let’s round out the list of what a lead magnet is not.

We’ll make the two premises above the starting point of this list, then add some even more important considerations:

What a lead magnet is not

• Something easy to make
• A list-building tool
• Something that you can’t sell on its own
• Something unrelated to your solution
• Something another business could create

The three points I tacked on are even more important than the first two.

Their connecting thread is this: a lead magnet is a product; your product.

It may not have the same scope as one of your paid products, but it must still have product-like impact and value creation potential.

When you create a lead magnet, think like a product manager, not a marketer.

Or like a drug dealer – does a drug dealer give away something the potential customer will simply throw away? No; they give something that is at least as good as the real thing, if not better. It’s only the quantity that’s limited.

BTW, there’s a finer point there; consider whether a lead magnet pertains to a single product – not to just any offer (product or service) your business provides.

So we design the lead magnet as a truncated version of whatever product it’s meant to inform people of,  interest people in, and sell people on.

In a sense, a lead magnet is an extension of the product itself.

A free product (or service) whose purpose is to make leads consider purchasing an related product or service that creates even more value.

(What it is not: a way to build a mailing list)

It’s been a while since I’ve posted here as I have occupied myself with building call transcriptions into Message Maps and posting over at The Microdose of AI.

With that, I wanted to flag a post I wrote over there today in which I overview the 12 most important developments over late summer. Because I am trying to distill over a months’ worth of news and developments, it’s more of a macrodose than a micro one.

But I think it’ll be of interest to anyone doing product work, design, development, marketing, etc.

Putting together a news round-up gave me a chance to reflect  – we may be at peak GenAI hype, but that doesn’t mean we’re in a bubble.

Tie that together with where we’re at on the calendar, near the start of Blair Enns’ proverbial 100 day sprint (from Labor Day to Christmas), the fall’s opportunity is to get better at weaving AI into our daily work.

You might find some ideas for how to do that over at the Microdose of AI – The most important AI developments over late summer.

A compound name derived from portmanteau’ing “rant” and “banter”. Long-winded comments, posts, or recordings with the energy of a rant but with more enthusiasm than opposition or bitterness.

TLDR – Art of Message will now be a weekly email; the new Microdose of AI list will be daily.

A month or so ago, I wrote about the (clumsy) misuse of AI as a deterministic software tool by the (suddenly archaic?) plagiarism detection industry. I say ‘misuse’ because AI works better as an adaptive, non-deterministic interface; it bridges the gap between desires encoded in messy human language and deterministic software/data tools.

Then a week ago, OpenAI threw in the towel here – quietly removing their AI detection “tool” (which was actually just ChatGPT on a different web page, to be honest) because it basically didn’t work.

*    *    *

Like anything related to Generative AI, this new item has a lot of implications for product strategy, no matter who you are.

But apart from it being a questionable PR move by OpenAI to sweep this under the rug, it has fewer implications for messaging strategy.

On that note, a change of direction for this list: Art of Message will go from being a daily publication on product strategy and messaging to a weekly one – with the same focus.

So I will see you here next week with my usual thoughts on product messaging, marketing, and strategy.

Meanwhile, I have started another daily list that will be more technical and more news-oriented: Microdose of AI 😛.

Like Art of Message, Microdose of IA is a “sawdust project” for me. Because I’m designing, developing, deploying, marketing, and selling a software tool that uses AI (and advising on several others), I follow Gen AI news daily – and have done so for more than a year. By coupling that firehose-info-intake with actually building things with Generative AI, I have developed a sense of what’s important in AI news, what’s trivial, and what’s just hype.

That’s what I’m sharing on the new list: clarity and perspective on the most important Generative AI news, technology, research, and stories of the day.

If you know anyone interested in that, please let them know – thank you!

https://microdoseofai.com/

It actually surprises me when people describe AI as having IQ. For example,  the “tests” which found that Bing AI (GPT4) has an IQ of 114 – haha!!

Not sure whether that’s just marketing, or a joke, or what, but I think it’s the wrong frame. Generative has an AI has an IQ of ‘null’ or ‘not applicable’, if anything. Same for EQ.

A child with an IQ of 50 (relative to an adult) can’t defeat Google’s 70 million dollar chess AI, AlphaZero, but they can do something much more intelligent: decide whether or not the game is to be played in the first place.

A better frame for AI-based automation is maybe UQ, usefulness quotient.

The first “UQ technology” I experienced was the automatic door opener at the Safeway grocery store. It had zero IQ but it was massively useful millions of times a day.

Over time, though, it receded into the background of shopper consciousness, taken for granted. This is nice because if users notice something too much, it actually becomes a little bit less useful.

I want the same thing for software built using AI, automatic usefulness that users hardly notice after a few experiences with it.

What happens if you accept that the true North Star is not a metric but an obvious and yet unquantifiable truth? What happens when you forget KPIs, analytics, conversions, goals, targets, and metrics? At least temporarily?

It gives you space to ask:

How do you provide overwhelming value relative to the cost of doing business with you?

I think that question is important enough to leave alone for now. Let me know if you have any thoughts; I will expand on new ways to do this next week.

Have a great weekend.