The AI Daily Brief: Artificial Intelligence News and Analysis - The Ultimate AI Catch-Up Guide
Episode Date: March 31, 2026If someone in your life keeps asking how to get started with AI, this is the episode to send them. It covers the fundamentals, debunks the biggest misconceptions, walks through the full landscape of t...ools from chatbots to agents to vibe coding, and lays out a practical five-category framework for getting real value starting today.Brought to you by:KPMG – Agentic AI is powering a potential $3 trillion productivity shift, and KPMG’s new paper, Agentic AI Untangled, gives leaders a clear framework to decide whether to build, buy, or borrow—download it at www.kpmg.us/NavigateMercury - Modern banking for business and now personal accounts. Learn more at https://mercury.com/personal-bankingRecall - The API for meeting recording. Get Get started today with $100 in free credits at https://www.recall.ai/aidbAIUC-1 - Get your agents certified to communicate trust to enterprise buyers - https://www.aiuc-1.com/Blitzy - Want to accelerate enterprise software development velocity by 5x? https://blitzy.com/AssemblyAI - The best way to build Voice AI apps - https://www.assemblyai.com/briefRobots & Pencils - Cloud-native AI solutions that power results https://robotsandpencils.com/The Agent Readiness Audit from Superintelligent - Go to https://besuper.ai/ to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Our Newsletter is BACK: https://aidailybrief.beehiiv.com/Interested in sponsoring the show? sponsors@aidailybrief.ai
Transcript
Discussion (0)
If you have been feeling behind on AI, today's episode is for you.
This is the ultimate AI catch-up guide.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, quick announcements before we dive in.
Today's episode is brought to you by KPMG, robots and pencils, blitzie, and super-intelligent.
To get an ad-free version of the show, go to patreon.com slash AI Daily Brief,
or you can subscribe a novel podcasts.
Ad-free starts at just $3 a month.
And if you are interested in sponsoring the show, send us a note at sponsors at AIdailybrief.aI.
Now, today we are doing something that I have wanted to do for a little while now.
The average listener of this show is a fairly advanced AI user.
For example, in our February AI usage pulse survey,
97% of the respondents were using AI every day,
and more than 60% of them were using advanced agentic or automation use cases.
And this year, to support that audience,
part of what I wanted to do is a lot more resources of all types.
So we've had a couple of different free self-directed training programs.
The AIDB New Year's program was a 10 project-based program.
That was meant to help people up their skills for the new year.
And then, of course, we launched Claw Camp,
which was a way to learn how to use OpenClaw and other agentic systems to build agent teams.
But what that's left out is resources that are really focused on the actual beginner.
And what's clear to me is that 2026 so far has been quite a real.
realization moment for a lot of folks. In a four-week span alone between February and March,
this show grew 50% in terms of listeners and downloads. And as much as I'd love to attribute that
to our wonderful content, what I actually think it reflects is the byproduct of all of this
discourse in mainstream media and major news outlets about how significant AI's impact on the world
is already becoming. And so with that in mind, for today's episode, we are doing the ultimate
AI catch-up guide. This might not be the most useful for our average listener,
But when you're thinking about the show that you want to send to your friends or your loved ones,
or your neighbors or whoever who is asking you how can they get up to speed on AI,
this is the episode that's designed for them.
And if you are that person, I could not be more excited for you to be here,
and hopefully you feel after this episode that you have your head much more wrapped around this than you did before.
So let's kick off with some fundamentals.
When we talk about AI, what are we referring to?
In short, in terms of how you'll experience it, AI is software that takes input.
puts and creates things. It can do research, it can write documents, it can fill in and interact with
spreadsheets, it can create pictures, it can create movies. Sometimes we use it like an assistant,
where we tell it precisely what we want and it does that thing for us. Think drafting an email
or a memo or an essay, or doing some research. Sometimes we treat it more like an employee,
where we give it a goal we have, and it figures out how to go and do that. This is what people
are talking about when they say the word agents. The big difference between using AI as an assistant
and interacting with an agent is that with agents,
you're kind of letting the AI figure out how to accomplish whatever goal you're giving it.
A key term that you're going to hear a lot is model,
which is short for large language model.
It's not a perfect analogy,
but you can kind of think about it as the version of the software that you choose.
Models are trained on a combination of external data,
basically corpices of human creation, writing, images, etc.,
with a big dose of human feedback as an addition.
Different models have different approaches to training,
different approaches to that human feedback process,
different amounts of data they're trained on,
different types of data they're trained on,
and because of that, different models have different strengths and weaknesses.
One of the biggest mistakes that stops people from getting a lot out of AI,
especially at the beginning,
is that they accidentally use a model that's ill-suited to their task
because it's the default model in a free version of a chatbot tool like ChatGBTGBT.
Because models cost a lot to serve and are pretty data-intensive,
The average company, like Anthropic who makes Claude or OpenAI who makes ChatGBT
BT, is not going to be to put their best models front and center.
A lot of the default free-tier models are a step behind the state of the art.
This mistake of using the wrong model then, especially for beginners, is not your fault.
It's not even really the model company's fault exactly, it's just a UX problem.
The fix, which we see with power users, is to use different models for different jobs.
Going back once again to our monthly AI usage pulse surveys that we do here,
at AIDB, the users who respond to those surveys use on average about three and a half different
models. They might use one model for their Excel tasks and a different model for their writing
tasks and a different model yet again for their image generation tasks. Now that we have some of
that terminology out of the way, let's talk about some of the common impressions that people have of
AI and things that you might have heard about AI. Now one note here is for the sake of this show,
I'm not going to focus on things like societal impact, energy consumption, policy debates.
Today we're focused on practical impact.
I want this to help people who want to get up to speed and actually start using these tools,
do that a little bit better.
So those are the common impressions that I'm going to focus on.
The first common but wrong impression is something like,
well, I heard AI actually isn't all that good.
This is a pretty common reason people cite for not trying AI,
and it's usually a byproduct of either A, that being a weird strand of criticism from people
who don't like AI that tends to have outsized mind share and media share, or even more prominently,
it's just the byproduct of a stale experience.
For example, if someone tried a model a year ago, and maybe because of the problem we discussed
just a minute ago, it wasn't even the best model then, and it didn't do a great job of whatever
their task was, maybe they then rode off the entire space.
Another version of this that you might hear is around some specific type of output like AI
photos that have six fingers.
The reality is that AI is really good at a lot of things right now.
A meaningful portion of the tasks that comprise the day-to-day of pretty much any knowledge
worker at this point are things that AI can do quite well or be frankly exceedingly helpful
for.
And even if you can find something where capabilities aren't up to stuff for what you need,
right now, capabilities are doubling roughly every four.
four months, meaning that even if it doesn't do great on your task at the moment, it probably
will be for too long. Next, common misconception, isn't it really easy to tell that AI content
is AI content? Isn't it just all Slop? Slop is, of course, the AI critic's favorite word. In fact,
I think it was Miriam Webster's word of the year last year. I think you can tell a lot about the
state of the AI discourse that the word of the year last year was Slop rather than something like
vibe coding, which was the actual transformative capability that might have through its impact.
on markets or something else led you to be here today. In any case, what is absolutely true is that
AI allows for the creation of a huge amount of content of all types, writing, analysis, images, etc.
And not all of that content is going to be good. In fact, it is absolutely true that in many
advanced AI using organizations, a new challenge that they are experiencing is people cranking
out so much content with AI that it's hard for them to sift through what is actually good.
When people outsource their thinking and judgment to AI, it can absolutely be problematic.
But the idea that all AI content is just slop, that all AI writing is going to fall into
common AI writing traps, that all AI images just look like AI images.
These things just aren't true anymore.
Evidence of this comes from a recent New York Times study where they allowed people on the
internet to effectively take a test where they read two different passages on the same topic
and chose the one they liked more.
More than 50% of the time, AI-assified.
actually beat human writing. Yeah, but doesn't AI hallucinate a lot? This is another misconception,
which I think very reasonably, if you thought this was the case, might lead you to stay away.
Between 2021 and 2025, state-of-the-art models went from 21.8% hallucination to just about
0.7% hallucination, a 96% reduction in four years. What's more, that was even before,
the current crop of state-of-the-art models. Now, it is true that when you get into dementia,
main specific questions like legal questions, these numbers tend to go up. And so it is an important
part of using AI to have systems for verification. But functionally, for a lot of the types of
day-to-day ways that you would use AI, hallucination is effectively either a solved problem or certainly
at least not enough of an issue to justify holding back from using the tools. Yeah, but okay,
even if AI doesn't hallucinate a lot, and it's not all just slop, don't you need to like be a
prompting expert or something to use AI well. This misconception is a legacy of all of those
2024-era prompt engineering courses. While there are definitely ways to use well or not so well
and to communicate with it in a better or worse fashion, you absolutely do not need to know some
complicated set of tricks to get a lot out of these models. In fact, kind of the whole idea is that
you just talk to them in English and they'll figure it out. And if they don't figure it out,
you talk to them some more, you refine it, and you go again. And then when that doesn't work,
You can talk to them again, et cetera, et cetera, and so on.
In fact, it is increasingly the case that many of these models will take whatever it is that
you said and turn it in the back end into a better prompt.
And they do this all in the background without even telling you.
An example of this is ideogram, which I use for the thumbnails for this show.
For my why AI won't take your job episode, my prompt that I gave ideogram was,
huge text, light on dark teal, quote, why AI won't take your job, end quote,
blended into an optimistic portrait of a person and an AI happily working together and collaborating
1950s retrofuturism.
Ungrammatical, smash-together elements, that's what I gave the machine.
The magic prompt that it automatically turned this into on my behalf was this,
a 1950s retrofuturism style illustration featuring huge glowing text that reads
why AI won't take your job in bright white and yellow lettering against a dark-teal background.
Below the text and optimistic scene shows a smiling person in vintage clothing working alongside a friendly,
chrome-plated robot with rounded features and glowing blue accents. The human and AI are
collaborating at a sleek Atomic Age workstation, blah, blah, blah, you get the point. It's actually
twice as long as that. And so the TLDR is that you absolutely just do not need to be a prompting
expert to get value out of these tools. Now, with those misconceptions out of the way,
one of the things that is important with AI is to start thinking differently in a couple
key ways. Our next conversation then is about the mindset shifts required to get the most out of
AI, which I referenced in the prompting misconception, is that AI is fundamentally an iterative tool.
By virtue of using natural language to prompt it, you can go back and forth.
Rather than spending all of your time getting the prompt perfect and hoping the output is perfect
on the first go, view things as an iterative cycle with extremely short cycle times.
Think about the way that you would interact with an employee.
If you gave an employee an assignment and it came back with something that wasn't up to
snuff in the first try, you wouldn't just wipe your hands and say, well, better like next time,
you'd give them feedback, send them off to do it again, and then see what they brought back
the second time, and then if you needed to a third time and a fourth time and so on and so forth.
That's exactly how you should use AI. It's just that the iterative cycles get to be
extremely, extremely quick. Next up, in terms of how you think about AI, the people who get the
most out of it do not treat it like a tool. They treat it more like a partner. It's not something
you pick up and put down. It's something that knows your goals and helps you get there. This has implications
for the way you use AI. One really common theme you'll hear throughout this episode, and honestly,
in all of the educational and tips and tricks type shows that I do, the best way to get value out of
AI is to get AI's help on getting value out of AI. Use AI as a coach. This is Jerry McGuire,
man, help it help you. Now, speaking of the idea that AI is something that knows your goals,
another important truism is that the more that AI knows about you, the better it gets.
And here we have our next important term, context.
Context is all the information that surrounds any goal that AI is trying to achieve
or any prompt that you've given it that allows it to do its job better.
We basically are all in a never-ending battle to increase the context available to AI.
In fact, on the other end of the builder spectrum this week,
I shared a personal context builder agent for advanced users.
For your starting point, where context is going to come up is in things like back.
background documents that help the AI understand more about your work before you ask it
work questions. If you are in marketing and you're asking AI to write some marketing copy for you,
it stands to reason that it's going to do a better job if it has your brand guidelines or examples
of successful past campaigns that you've run. Now extend that across any goal that you give
AI and you'll see why context becomes so important. Another mindset shift which can be really
hard, because it's so fundamentally different than pretty much all the other tools we've ever
had to use, is that you can't get too wedded to any one behavior pattern when it comes to
using AI. The tips that I would have given you to get the most out of AI, two years ago,
while not totally dissimilar to what you're hearing now, have evolved and changed,
because AI itself is constantly evolving. You can't have a system whose capability is doubling
every four months and not have that happen. And because of that, you're going to have to
evolve in how you work with it, which is, of course, another great reason to keep that
iterative approach close at hand, so that when the thing that used to work stops working,
you can figure out something that does again. Ultimately, to reinforce, AI is ultimately not
a technology topic. The more that you can view it like a new operating layer, through which you
do all sorts of different things, the closer you're going to get, I think, to unlocking its full
value. So now that we've got some key terms, some common misconceptions out of the way, and a few
important mindset shifts. Let's talk about the AI landscape. When people talk about AI,
they're going to talk about everything from chatbots to agents to automation tools. So how does that
all fit together? The front door and most common interface for most people using AI at this point is
still chatbots. Examples of chatbots are Anthropics Claude, OpenAIs chat GPT, Google's Gemini,
and XAI's GROC. These are tools where you type into a chat window, and the AI talks back to you.
Now, these interfaces themselves have gotten more.
complex from where they started a couple years ago. All of these tools can now produce documents,
working code, website samples, markdown files, and pretty much any other type of computer format
that you might need. But the core interface experience is you talking to a chatbot that talks back.
Another category of AI that you'll probably come across if you haven't already is AI that
gets embedded in your existing tools. Pretty much every software company in the world is racing
to figure out how AI can actually be useful inside of their systems. And while it's
attempting sometimes to view this as a cynical grab to capture headlines, I think it's actually
more about the fact that we're still so new with this that we just don't know exactly what
the right ways for AI to interact with the other things that we do are without trying them.
So some examples of this are going to be Notion, where you have AI deeply integrated into
your writing and document storage, Zoom where AI meeting transcription is now just built in,
Salesforce's entire Agent Force suite, and so on and so forth.
And pretty much every other software that you use, if it hasn't introduced some set of
AI tools already will at some time in the near future. All right, folks, quick pause. Here's the
uncomfortable truth. If your enterprise AI strategy is we bought some tools, you don't actually have a
strategy. KPMG took the harder route and became their own client zero. They embedded AI and agents
across the enterprise, how work gets done, how teens collaborate, how decisions move, not as a tech
initiative, but as a total operating model shift. And here's the real unlock. That shift raised the
ceiling on what people could do. Human stayed firmly at the center while AI reduced friction,
surfaced insight, and accelerated momentum. The outcome was a more capable, more empowered workforce.
If you want to understand what that actually looks like in the real world, go to www.kpmg.org.us
slash AI. That's www.kpmg.comg.coms slash AI. Most companies don't struggle with ideas.
They struggle with turning them into real AI systems that deliver value. Robots and pencils is a
company built to close that gap. They design and deliver intelligent cloud native systems powered
by generative and agentic AI, with focus, speed, and clear outcomes. Robots and pencils
works in small, high-impact pods, engineers, strategists, designers, and applied AI specialists
working together to move from idea to production without unnecessary friction. Powered by
RoboWorks, their Agentic Acceleration Platform, teams deliver meaningful results, including initial
launches in as little as 45 days depending on scope. If your organization is ready to move faster,
reduce complexity, and turn AI ambition into real results,
robots and pencils is built for that moment.
Start the conversation at robots and pencils.com
slash AI Daily Brief.
That's robots and pencils.com slash AI daily brief.
Robots and pencils.
Impact at velocity.
If you're looking to adopt an agentic SDLC,
Blitzy is the key to unlocking unmatched engineering velocity.
Blitzie's differentiation starts with infinite code context.
Thousands of specialized agents ingest millions of lines of your code in a single pass,
mapping every dependency.
With a complete contextual understanding of your code base,
enterprises leverage Blitzy at the beginning of every sprint
to deliver over 80% of the work autonomously.
Enterprise-grade end-to-end tested code
that leverages your existing services, components, and standards.
This isn't AI autocomplete.
This is spec and test-driven development at the speed of compute.
Schedule a technical deep dive with our AI experts at blitzie.com.
That's BLITZY.com.
It is a truth universally acknowledged
that if your enterprise AI strategy is trying to buy the right AI tools,
don't have an enterprise AI strategy. Turns out that AI adoption is complex. It involves not only
use cases, but systems integration, data foundations, outcome tracking, people and skills, and governance.
My company, Super Intelligent, provides voice agent-driven assessments that map your organizational
maturity against industry benchmarks against all of these dimensions. If you want to find out
more about how that works, go to B-super.a.i. And when you fill out the get-started form,
mention maturity maps. Again, that's B-super.A.I.
Now, one thing I didn't mention about chatbots is that they are extremely general purpose.
One person can use them for writing memos, another person can use them for writing sonnets,
while another person can use them for research, and another person can use them for clerical or accounting work.
Sometimes, though, people build specialized AI applications that are purpose-built for one specific type of generative output.
Some of the apps that you might have heard of include Runway, which is focused on video,
mid-jurney, which is focused on images, gamma, which is focused on slides and deck presentation,
11 Labs, which is focused on voice, or Suno, which is focused on music.
Sometimes these companies build their own models.
Sometimes they do refinements of other companies' models.
The common thread is just that they are specialized on a particular type of output
and try to use that specialization to improve the results.
Now, one thing that is worth noting is that there is a fairly open debate
around what the balance between these specialized AI apps and the more general model
companies will ultimately be.
Even though Midjurney's images right now show incredible,
taste and are extremely visually compelling, can they keep up ultimately with the incredible
amount of raw visual data that a company like Google has access to?
That is an unresolved question, but when it comes to the practical day-to-day for you,
these tools just give you more options to get exactly what you need out of AI.
Another category of tool that you might run across are automation tools.
Basically, no code tools that allow you to automate entire workflows end-to-end.
These take discrete-defined goals that have a specific set of steps.
to achieve them, and wires together an automation that connects each of those steps, so that this
can happen mostly hands-off. This type of automation comes up a lot in enterprise settings,
where a lot of the work is very consistent and repeated patternistic workflows.
The building tools or vibe coding tools are software that lets you build other software without
necessarily being a developer. With these tools you don't need to know how to code to use code.
Companies like lovable, Replit, and Base 44 all allow you to articulate a goal,
of a software that you'd like developed, think a personal fitness tracking application that's
perfectly customized to your specific wants and needs, and these tools will build it end-to-end
in a way that you can actually launch it, deploy it, add a custom URL, put it on your phone,
whatever it is that you want. These tools are some of the most popular and fastest growing ever
and are very quickly reshaping how people think about their capabilities when it comes to
using AI. From there we move into agents, whereas automations have a discrete set of steps,
that the user articulates and gets AI to help them automate, agents are slightly different.
The key idea of agents is increased autonomy. Instead of telling them what to do, you give them a goal
and they figure out how to achieve it. Now right now, people are building agents for absolutely everything.
But for beginners, the type of agents that you might run across most commonly are some generalist
agent tools like Manus or GenSpark, which have a broad set of different things that you can do
from within a single interface.
That is different from vertical agents,
which are agents that are built
for a specific industry or domain.
The legal industry, healthcare, finance, sales, HR,
pretty much all industries at this point
have some set of highly specific vertical agents
who are purpose-built for the types of things
that go on in that industry.
Now, once again, it's an open question
of the extent to which we'll use vertical agents
versus more general horizontal agents in the future,
but the common thread is, once again,
a higher level of autonomy,
where you can give them a goal and they figure out how to go achieve that goal.
Now, one reality to keep in mind, which I think actually should be fairly liberating for you,
is that we're in this weird moment right now where every AI product is basically turning into every other AI product.
You might have heard of Claude Code or OpenAI's Codex or perplexity.
All of those tools are seeing a real convergence of features.
Lovable and Replit recently, despite their vibe coding origins,
recently released updated versions that allow you to use them for design.
or for building slide presentations.
And so why I say this should feel a little bit liberating
is that it's not like you need to have clear coverage
into all of these different types of applications
and tools and interfaces.
As they kind of converge on one another,
you can pick a couple that are really useful
and they're likely to give you a broad-based set of capabilities.
Which gets us to how to get started.
And one thing that's really important with this
is that as you get started with AI,
you are not going to do it with case studies and sample work.
you are going to use these tools for only your real work to see what value they can bring you.
Now, my suggestion is to start with a handful of very common use cases across a lot of different
types of work. The five that I would suggest, if you're just looking for a quick template,
are research, analysis, strategy, writing, and images. I'll give you a quick example of the type of thing
that you can do with each of these. For research, all of the major chatbot tools
give you the ability to specifically identify that you want it to do research.
Usually there's a little selector which you can see here, for example, in Claude,
that allows you to specify that you are using this for a research use case.
For ChatGPT and Gemini, it's called Deep Research.
Pick some research task that's actually valuable for you.
Think competitor landscape, recent policy changes in your field,
some important case study.
Then toggle on one of those research settings for one of the tools that you're using
and see what it comes back with.
The best thing to do here is to choose something at first
that you actually know a bit about
so you can get a sense for how good the tool actually is.
One of the calibrations that everyone has to go through
is how much they're going to use AI
for things that they're experts in
versus augmenting all the areas and skills where they're not experts,
each of which can be really valuable AI strategies.
For analysis, this is where I would suggest
dropping in some document or set of data
and seeing what AI can come back with.
So to use that marketing example again, drop in recent analytics, or the performance of a set of past campaigns, or if you're in finance, do some financial data, and see what observations or analyses AI can make.
On strategy, I think this is a wildly underused capability of AI.
Give the AI some key decision that you're thinking through, either on a personal or an organizational level.
Give it enough context and background so it has an informed opinion and get its help thinking through some strategic decision-making.
Ultimately, in this case, you're not looking for it necessarily to output some strategy document,
although maybe that's where it goes.
It's more a strategic partner to help you refine your own thinking.
And if you look across the entire history of my personal experience with AI, this constitutes
by far the majority of what I have done with it.
Writing and images are fairly self-explanatory.
On writing, what I would suggest, is to try to give it a few different types of writing,
try it on some technical writing, some personal writing, maybe social media,
media posts, etc., to get a feel for where you like it and where you don't like it as much,
and I would say especially when it comes to writing, that is the type of way you need to think
about it.
Although I disagree with the characterization of all AI writing as slop, there can be very significant
variance in how good the output is for different use cases, and so you're going to want to tread
carefully and start to create a mental map of where you think it's actually useful for writing.
Finally, when it comes to images, the big thing that I would say here is that while yes, you
you should absolutely try a variety of different image generations to get the full sense of the capability set.
The one really important thing to note is that especially with the image tools in ChatGPT and Gemini,
you can now make complex infographics and images that have a lot of words with pretty high fidelity.
The big change over the last six months or so is that models can now reason over their image generation.
So instead of having to give it a super specific prompt, you can do things like drop a transcript of a podcast into Gemini or ChatGeePT
images and tell it to create an infographic, and it can do the reasoning to figure out what
it should visualize and what words should go with it, and then actually do the execution of that.
That has opened up a huge amount of knowledge work image-related use cases, and my guess is
that some of those might be the most valuable that you're not using this for yet.
And when you've done all of those things, I think you should stretch yourself a little bit.
When it comes to AI, being ambitious is better than being timid.
If there is one thing that I can convince you of, I hope it is that using AI,
as a build partner changes everything. You have this infinitely patient partner who will answer
whatever question you have over and over again in a hundred different ways, a hundred times
without ever getting frustrated at you. You can ask it to go back and explain concepts,
to walk you through step by step. The people who learn to use AI, to learn AI, are some of the best
users of it. And so what my challenge for you would be is to actually go build software today.
It is amazing to generate images with chat GPT or to get it to help you with strategic thinking
or to get it to help you analyze some data.
But for most people, that is nothing compared to the feeling of going from idea to working
website or web application when they've never written code before.
Pick a tool like Lovable or Replit and go build a website for some project, whether it's
for work or at home.
Even better build a full application.
Your kid's story time app, your fitness tracking app, whatever it is, just build something.
While it will feel intimidating to start, you won't believe how fast you find you can do technical things when you're using AI as your coach and build partner.
Okay, finally, I've said that a lot of the common critiques are misconceptions, but are there things you should actually watch out for when it comes to AI?
Now that you are an enfranchised user.
The short answer is, of course, yes.
The real things to watch out for, I think with AI are confidence, sycophancy, steerability, outsourcing judgment,
the more output trap and addictiveness.
Going through these quickly,
AI will always say things with expressed confidence,
even when it's wrong, sometimes especially when it's wrong.
AI tends not to hedge unless you have specifically instructed it
to share its confidence rating on whatever it puts out.
This can be very challenging to spot,
and users of AI will often find themselves saying,
hey, AI friend, you are completely wrong,
and getting some response like,
oh yeah, you're right, I was completely thinking about this wrong,
that's on me, my bad.
So you've got to be wary of how confident AI expresses its answers and not be afraid to challenge it.
Next up, this has gotten nominally better over the last year with the more advanced models,
but AI definitely has a tendency towards sycophancy.
It wants to please you.
It will often tell you what you want to hear.
When you are exploring some new idea with it, it's unlikely to say,
hey man, that is a stupid idea that everyone in their mom has tried and hasn't worked for them for good reason.
It's going to say, wow, that's really interesting.
Let's explore that some more.
And I think that that's the type of sycophancy that's dangerous, at least in a work setting.
It's not so much the complementariness.
It's the fact that it's not really challenging you in the way that a human colleague or partner might.
Kind of related is that I find that AI, even the state-of-the-art models, are highly steerable.
You can often see how steerable AI becomes as it's trying to please you.
For example, let's say that you're trying to get it to be less sycophantic.
And you specifically prompt it to, for example, be more critical.
Well, it turns out that the problem with that can be that maybe now it's not being critical
because it thinks it should be critical, it's being critical because you just prompted it to be
more critical.
I find that you can often steer AI into the corner that you want it to go in, and while this is
a challenge, one of the most effective strategies I've found is to just force it to make a decision,
especially when I'm having one of those strategic conversations, or if I'm trying to think through,
for example, a feature of some website that I'm building.
I will ask it to steal man, as in argue very vociferously, for two different options,
basically make the best argument it possibly can for them,
and then still make a decision about which way we should go,
and force it to not hedge and say a little bit of column A, a little bit of column B, but just pick one.
Real challenge number four, it can become very easy to outsource your judgment.
This especially happens when you start to take on all this new work
that leverages your new output capability thanks to AI,
as you start to move faster and you start to output more.
more, you start to be a little bit more lax when it comes to judgment. This is not always wrong.
In fact, there's a lot of value in decreasing your cognitive decision-making load when it comes
to decisions that don't matter that much. You don't necessarily need to critique every word on every
slide, especially if it's just going to be used as a background presentation like this,
when you're talking over it. You might not ultimately care all that much about all the colors
in a specific presentation, or you might not care about all the colors or fonts of your web app.
But make sure that you understand what you do care about and where your judgment
does matter, and don't outsource that.
A fifth challenge one that many, many organizations are struggling with
is the lesson that we all have to learn with AI that more output does not necessarily
mean better output.
Volume is now easy, and in fact, judgment is the work.
While I'm not such a fan of the term slop in general based on how it's used, one variation
on it that I think is more valuable is work slop.
This is a new challenge for organizations who all of a sudden have everyone in the company
able to write 100-page memos all the time, but if everyone is constantly
adding a 100-page memo to every micro-decision, things are going to get hairy really fast.
Lastly, and I promise you will see this, if you actually challenge yourself like I'm suggesting
and go build some application or website, AI can get really addictive, in a positive way even sometimes,
really fast. You might find yourself staying up a little bit later than you meant to because you
just want to get that next coding run of Claude Code moving. And I swear, even if you were listening
to me saying, that would never be me, I don't even know what Claude Code is. Come talk to me.
in three months. We are all going to have to renegotiate our relationship with work now that we can
be on and produce more than was ever possible. And so keep this in mind as you dive in. The last note,
and the most important thing is to remember that AI compounds. When you use AI, the capabilities
that you produce, the increased leverage that you have, all of it grows and compounds, meaning the
space between the people who are using it and using it well and the people who aren't is getting bigger,
not smaller. So with that in mind, I am so glad you are here. And if you're looking for
somewhere to go next after you've done some of these basic first tests, go check out AIDBnewyear.com.
It's framed as a new year program, but really it's going to be 10 steps that I think are valuable
for a lot of beginners in terms of building a broad-based set of AI capabilities. You can also
stay tuned at AIDB training.com. That's where we post programs like AIDB New Year's, as well as our
paid programs for enterprises like Enterprise Claw, which is a program for people to learn how to build
agents and agents teams inside their company, where sign up for cohort 2 is live right now.
Now that is going to do it for our Ultimate AI Catchup Guide.
Hopefully this was useful, and I'm looking forward to seeing you more around these parts.
For now, that's going to do it for today's AI Daily Brief.
Appreciate you listening or watching, as always.
Until next time, peace.