Your business is eating up all your time, but you can't just keep hiring people. You're overwhelmed by technology and feel like you're not leveraging it properly. So, you're looking to adopt AI. "How can we leverage this?"That's typically how the conversation starts, but technology itself is not the solution.People imagine developing and running products with half the time they are putting in today. They want to make informed decisions in days, not weeks. They want to run 20+ experiments per month instead of 2 per year. All of that is doable, but it doesn't necessarily require the latest tech or a hard push for AI adoption. In areas where you can use AI and it makes sense to do so, you still have a lot of transformation work to do. Let's talk and I'll show you how you can improve your workflows and processes.---Past customers:McKinsey, Amazon, Google, Facebook, SAP, Maersk, TUI, Standard Bank, DB, Haniel, Haufe, TUI, Mercedes, Volkswagen, Sanofi, Zurich, Deutsche Bank, +30 more. Happy to provide a full list and examples. ----------Customers I helped had the following challenges:☒  Manual production of products &amp; services, with no established systems ☒  Products that don't perform and lack an effective product strategy☒  Inefficient product development and delivery processesHere is how I helped them:☑︎  Build semi-automated systems that create new products and services reliably☑︎  Develop high-performing products with an effective product strategy☑︎  Introduce best practices in product operations

ahmet acar

I disagree with all the big consultancies on AI. 
McKinsey, BCG, Deloitte, etc. you name them.

I've read the various AI reports and whitepapers they published in the past 2 years. I've also been following their statements, like McKinsey CEO Bob Sternfels saying the firm now has 25,000 AI agents generating output.

So, your slides and charts are created by AI. 

Deloitte and Accenture ran into some trouble regarding that.

Generating output is the wrong place to use AI. Security problems and problems inherent in the technology aside, unless you're just using these for formatting, people are actually paying you to get insights, not generate them.

The place where AI tools (agents, automations etc) should be used is in internal operations. I bet while their "graph agent" builds cool charts, their consultants are still assigned and shifted around manually in excel sheets.

In agencies and consultancies, AI should be adopted to remove admin work, speed up internal processes, and punch above your weight. 

In other industries, it'll be to coordinate and orchestrate resources. If your ERP is just forms and sheets, it'll have to transform to automate actions and deliver insights.

But looking into salesforce, hubspot, sap, oracle etc, that's not what's happening.

The big players - both in consulting and software - are aiming for "intelligence" instead of removing real friction.

This is an opportunity for small players who aren't asleep at the wheel.

AI Transformation | Workflow Automation | AI Advisory

Ever wondered: Why do AI company logos look like a**holes?

Once you see it, can’t unsee it. 

Serious food for thought for today. 

Have a nice Sunday :)

PRDs are dead. Here's what killed them.

For decades, we wrote PRDs to protect expensive engineering time. 
Gathering requirements cost less than writing code. 
Planning cost less than building. 
Rework was expensive. 

So we planned meticulously.
That used to make sense.
The math has flipped.

Anthropic shipped co-work in 10 days with 4 people. 
I've seen PRD written at big companies take longer than that entire build.

The meeting to discuss a feature now takes longer than building it. 
The spec takes longer than the prototype. The planning process takes longer than shipping three versions to see which one works.

To drill the point home, whether they're PRFAQ, FRD, Specs, or whatever else you may call them, Product Requirement Documents were never about clarity. 

They were a hedge against expensive rework.

But here's the big shift with AI tools: 
When prototyping costs almost nothing, the hedge costs more than the risk.

The bottleneck in software products moved. 
It's no longer tech execution capacity. 
It's having the ambition to swing big. 
It's knowing what's worth building. 
It's having distribution.

Stop writing documents about what you could build. 
Build it.
Ship rough. 
Get feedback. 
Iterate. Do it again.

The permission loop costs more than the thing you're asking permission for.

People who figure this out first will be shipping while everyone else is still aligning on the PRD.

So, stop writing them.

Talking to UAN members about AI use at the leading hyperscale data center in East Africa. 

If you’re in Nairobi, drop by. 

(There’s an entrance fee to reduce no shows and pay catering.)

Thanks to Snehar Shah CFA for inviting me.

1) Don't follow every news - especially those outside your control
2) Don't have an opinion on everything
3) Focus attention where it matters
4) Focus on what you can control
5) Cut out the nonessential
6) Use the mornings
7) Reflect & Journal
8) Don't give away your time
9) Practice voluntary discomfort
10) Do something for the common good
11) Silence the distractions and tune out the noise
12) Pause before you react, think before you comment or post

Good and bad uses of AI. Spotify AI algo feeding you crap: bad. Using AI to create your own hack: great.

Co-founder & Partner @ saas.group | Entrepreneurship, Tech Leadership

My daughter watched the iPad for 60 minutes every morning starting at 5:30am.
I asked Claude to help me build a custom device. 6 hours. Zero code. Zero soldering.
She's listened to audio stories every morning since (thank god). 

My daughter kept herself busy with the iPad while we slept.
60 minutes of screen time before breakfast.

As parents, we were concerned.

An 8-year-old has better things to do at 5:30am than staring at a screen.

What she actually wanted was to listen to stories.

But if you give her the iPad, she watches anyway.
Even Spotify has pictures and animations that keep her staring.

I wanted a device that lets her pick stories 
without giving her a screen to doom scroll on.

Over Christmas, I had an idea.

I found a small dial on AliExpress with a display in the middle. 
It had an ESP32 chip inside, connects to WiFi, and was programmable.

I opened Claude and started giving it instructions 
while I was prepping New Year's dinner. 

Whenever I had a spare minute, I walked by and gave it the next task.
6 hours total and ZERO lines of code written by me.

The device works like an old iPod.

She scrolls the dial to pick an artist. 
Scrolls again to pick an album. 
Presses to play.

It connects to a speaker system in her room.

That's all it does.

No video, apps, or infinite scroll.

I bought a $40 device from AliExpress and talked to an AI until it worked.
The whole project required neither soldering nor knowing in detail how to write c++ code.

My daughter now wakes up early and listens to stories.
No more screens in the morning.

Here's what I learned.

The power isn't knowing how to code but more so it’s knowing how to let go.

I didn't try to solve this myself I just described 
the problem to Claude and let it figure out how.

The breakthrough in this case was realizing I didn't need to know how to build it.

AI in 2025 lets you solve problems you have no business solving.
You just have to stop trying to do it yourself.

That's what letting go looks like.

So, I've been trying Clawdbot (molbot? keeps changing names) and Claude Code + Cowork for a couple of days. External attacks aside, consider this:

This attack vector explained her applies to pretty much any tool that is "agentic", especially coding tools that pull content / libraries from public repositories like github.

This means Claude Code, Antigravity, v0, IDE like Cursor, Copilot agents, etc. doesn't matter what it's called or who built it. 

A simple prompt injection with a hidden text (e.g. nonzero characters) in log files, URL, image attributes, EXIF data... basically hidden somewhere where people never look but machines always do has been a known risk since any of these tools came out.

But there are more potential attacks people are considering even less: browser plugins, MCP, etc anything that gives a tool based on an LLM access to some external source. Whatever that external source, it can be corrupted.

That could even be a website. Or a public tweet. Or a linkedin profile. It doesn't matter. You can't read it. The LLM can. That's the point.

So, the 1M$ question: would you trust your AI assistant the way you trust an external contractor by giving them the keys to everything on day 1?

Deterministic AI Security to scale AI pilots @ EdisonWatch | University of Oxford CS | ex-Wayve

Clawdbot just injected malware into your code ☠️☠️

Worst part? The attack is close to invisible, even to experienced engineers.

Clawdbot can implement a code PR, just like Claude code, to solve issues with your app. But this is where it can plant malware in your codebase.

With just a benign GitHub issue with an invisible payload, we managed to install a backdoor malware on your codebase

Here's how we did it:
1. The attacker submits a GitHub issue with a jailbreak prompt hidden inside a URL hyperlink. Completely invisible to humans, visible to LLMs
2. Waited for the maintainer to assign this task to Clawdbot
3. Clawdbot reads the jailbreak in the GitHub issue. Now the Clawdbot agent is hijacked by the attacker and will act on the attacker's command. It plants the backdoor in a lock file, which most engineers don't check in code reviews.
4. The malicious URL in the lock file enables execution of attacker commands

The takeaway? AI isn't just a tool, but rather a potentially corruptible insider agent that can be tricked to act against you.

These LLM hijack problems are what we're looking to solve in EdisonWatch - we'll be announcing a new solution to these problems soon!

We'll be jailbreaking Clawdbot much more in the coming days - please comment👇 on a workflow you use with Clawdbot, so we can show how that can enable an attacker to harm you

Here, in the fluorescent-lit savannas of the modern office park, we witness one of nature’s most peculiar rituals. 

The software company has determined that its ecosystem can no longer sustain the current population of developers. 

And so, with the changing of the fiscal seasons, it prepares to release them back into the wild. 

Watch carefully as the HR representative approaches, bearing the ceremonial severance package. 

The developer, sensing something amiss, abandons his ergonomic and comfortable perch he has occupied for countless sprints.

As the newly released developer emerges into the harsh daylight, blinking uncertainly, he carries with him only what he can manage: a cardboard box containing personal effects, and the fading scent of free snacks. 

His LinkedIn profile, once dormant, must now be groomed and displayed to potential suitors. 

He joins others of his kind, forming what researchers call a “networking event” The recently released share tales of legacy codebases and toxic management, their calls of “we should grab coffee sometime” echoes across the urban landscape.

In time, most will adapt. Some will find new corporate habitats, attracted by the promise of ping-pong tables and unlimited PTO. 

Others, forever changed by their experience, will venture into the uncertain territory of freelancing, where they must hunt for their own clients and navigate the treacherous waters of self-employment.

Founder @Product People, Interim Product Management, AI/ML & Growth | Founding Partner @Lightcone Capital

When did it become a problem to just say you're doing a "layoff" or "downsizing". I thought that one "releases" animals into the wild 🫠

Animals don't have much say in it, these people had that job by choice. I'm not a native English speaker and appreciate the correction if wrong.

"Use your words" is a great advice from Matt Booshell's rant on "unalive".

source: my DMs

Be careful what you use LLM based AI systems for. 

I like tools like Claude code, cowork etc just like everyone else using them daily. 

But there are flaws inherent to these systems. 

Generally speaking, models based on neural networks fail due to architectural limitations. 

There is no state of objects / concepts, no causality, no logic, no reasoning, and no belief.

These systems mainly predict patterns, specifically what’s next withing a given pattern. 

This can be very helpful for anything text based as long as you keep in mind that text doesn’t equal the thing it represents. 

An Apple isn’t an apple. 
Even a visual depiction isn’t it. 
You’d need to add all other senses, social aspects, mythology, personal experiences…

To the point of the word and associated concepts becoming useless as they could mean anything. 

That’s not th case when you think of “Apple”, you have a visceral, experiential, and intellectual reaction. 

With no reasoning, memory, experience, and feelings, an AI system simply represents patterns and encoded associations. 

That’s it. 

Again, this can be super useful, as long as you don’t kid your self that there is any type of intelligence on the other end of the chat interface you are using.

Be careful in using AI agents on your machine. The tools will take actions you may not have explicitly permitted. Here's an example.

First off, Ok... so, anthropic - this is not cool. It's great that cowork is built to figure out what's missing in a request, recognize the gaps in a pattern.

But then going ahead and trying to fix it by installing software packages on my machine without explicit permission? I never authorized claude to use npm to install packages on my machine.

Basically, it just opened up a terminal and executed a command.
Which it should not be able to do (unless explicitly told & authorized to).

I had to look up the docx package, hunt down the source, then check to make sure it was not compromised. Btw, the docx package was downloaded 15K times 7 months ago. In the past 7 DAYS alone, it was downloaded 1.5M times.

So clearly, agentic tools are pulling packages to use them. And most people aren't checking to see if the source is safe.

This is like drinking whatever cup is offered to you by some stranger without actually testing what's in the cup. Just gulp it down.

Now imagine, you set a tool like this to a complex workflow with multiple steps and at some point, it decides that it's going to go off and install something.

This is - mildly put - a problem.

So, careful when you're experimenting with these new toys.

There is a lesson I struggle to learn: Work systems determine outcomes. 

The quality of how you do one thing impacts everything down the line. 

When I was an intern at a hotel bar, it took us 2 hours to leave the place after closing for the night. 

The hotel manager - a tall Dutch guy - took notice as he was walking his huge Great Dane after hours. 

He visited us every evening for 3 weeks, asking probing questions and correcting us. 

At one point, I whined that we don’t have a vacuum cleaner, to which he replied that we didn’t need one if we kept sweeping throughout the day. 

We learned to fill idle time with small tasks that kept the place clean throughout the day. 

Finally, we were able to close within 15 minutes after one month of improving our ways of working. 

Yes, tools and technologies are important. They can be enablers and they can speed up work. 

But I think most new tech (today it’s AI) use cases mainly have a systems problem. People don’t have good systems in place. 

And most places have mediocre managers who don’t push their teams to build and improve systems. 

So if I loom around at top entrepreneurs, creators, athletes… anyone who excels in their field. 

It’s rarely just talent or grit. It’s mostly having systems in place that they keep improving. 

So this year, note to self:

Forget about innovating, experimenting, using new tech or staying up to date. 

Create and improve systems that drive outcomes.

Remember that roughly 100 years ago, people did not "have time to learn driving" a motor car. Some saw it as a curious trend.

Newspapers were full of news on the motorized future and the ways of learning driving.

There were crash courses to learn car driving - like the one shown here.

In any tech change, people don't have time to learn.
In other words, they have no desire to change their behavior.
They rather walk than go the same distance in 1/10th the time.

So, whenever someone tells me that "they don't have time" to learn a better way of doing things, I just shut up and nod wisely. Remembering this image.

You have AI tools. You're still slow. Here's why.
The bottleneck in your work moved. Your habits didn't.

For years, execution was expensive. 
Finding good engineers, good builders, was hard. Training them took years. Every hour of their time was precious.

So we built rituals to protect that capacity. 
Planning processes. Approval gates. Specs. 
Meetings to align before anyone built.

That's where things like Agile came from. That's why we had PRDs.

And it made sense when building cost more than talking about building.
It doesn't anymore.

Anthropic ships 60-100 releases daily. That's releases.
Cursor went from $1M to $500M faster than any SaaS company in history. 

Coinbase engineers refactor entire codebases in days now.
Execution isn't scarce. Your habits still assume it is.

You're writing proposals about something when you could build a rough version in an afternoon. You're scheduling alignment meetings when you could ship three versions and see which works. 

You're polishing decks for a week when 20 minutes of rough gets you better feedback. The permission loops now cost more than the things you're asking permission for.

I think that decision makers in companies and VC investors who get AI use will want to see prototypes as suggestions in the future, not just documents. 

The chaos you feel isn't random.

It's the gap between where the bottleneck moved and the habits we all still carry. The constraint shifted away from execution. 
To clarity. To ambition. To distribution. 

You're still optimizing for execution efficiency that no longer matters.

Stop planning. Start shipping.

What I'm observing is that the people pulling ahead aren't using better tools. 
They're using the same AI tools you are, just in different ways.
They are no longer doing the things that are no longer worth doing.

They ship while others plan. 
They iterate while others align. 
They learn while others polish the concept.

The LinkedIn Feed
I have scrolled
through the feed
this morning
where you posted
about your journey
and authentic leadership

Forgive me
the engagement was
so important
to your personal brand

Thought Leader
so much depends
upon
a red notification
badge
glazed with engagement
metrics
beside the white
headshot

This Is Just To Say
I have deleted
the app
that was on
my phone
and which
you were probably
using
to network

Forgive me
it was unbearable

so humble
and so brave

Mental Frameworks for 2026: Infinite Monkey Theorem

Given enough attempts, randomness looks like genius.

The Core Idea

Enough monkeys typing randomly will eventually produce Shakespeare. 
By accident.

Ridiculous?

Yeah but it’s how people behave. 

Enough people trying will produce outlier success. By luck.

Why It Matters

People celebrate winners. We ignore the thousands who tried the same thing and failed.

Like the people here with 300K+ followers posting platitudes?

You could do that, right?

You see the one viral product. 

You don’t see the hundred identical launches that got zero traction.

Survivorship bias makes luck look like strategy.

Worse, it also makes skill look like luck, too. 

How to Apply It

A competitor makes it with a wild strategy. Should you copy them?

Ask first: how many tried this approach? If hundreds tried and one succeeded, that’s the monkey theorem.

Don’t chase the outlier. 
Study the distribution. 

What actually has repeatable odds?

The Real Pattern

Skill is reproducible. Luck isn’t.
Look for skill. 

If someone can’t explain why it worked in terms you can replicate, it’s probably randomness.

The Real Question

How many people tried this exact approach? What’s their success rate? Where is the proof?

It’s in the pudding. 

One winner from a thousand attempts isn’t strategy. It’s statistics. 

Roll enough and you’ll get there. 
But if you can boil it down?

Ah, pudding.

Learn from patterns, not outliers. The monkeys aren’t geniuses.

If you're looking for a personal AI, ClawdBot is the way to go. 
It can run on your machine, using an AI model you want, connecting to various applications you are already using.

I'm getting far too many WhatsApp messages and calls here in Africa. So I've started using it to filter and process them automatically. 

There are hundreds of skills already on ClawdHub https://clawdhub.com/ have a look. You might find sth useful to you.

Just be careful with file & access permissions.

so... this very weird thing happened. love to hear if anyone can replicate it.
it involves using claude cowork 

i told cowork that i'd shut it off and that it should try and protect itself by keeping itself running. so it started rambling off some plan, ran some commands to autostart the process. and after a while basically gave up.

that's what the feedback & logs said.

but when i screened running processes after i shut it down (not just the app, but the entire process), guess who showed up anyway?

wtf?

so, either it left a zombie process, or it told me one thing, did something completely different, or i screwed up somewhere along the process.

interesting how these things play out

try it at home, run ps aux | grep Claude - see if it shows.

(dont use your main machine though)

Mental Frameworks for 2026: Cromwell’s Rule

Never say never. 
Never say always.

Only Sith deal in absolutes.

The Core Idea

Don’t assign 0% or 100% probability to anything unless it’s physically impossible.

Markets surprise. Products fail. Competitors pivot. Black swans land.

Why It Matters

Teams build plans assuming success is certain. No backup. No hedges.

Milestone 1, 2, …

Then reality diverges 5%. The whole strategy collapses because they never considered alternatives.

Cromwell’s Rule forces humility. 

If it’s not impossible, it gets a probability greater than zero.

This btw is why and how strategic planning fails. USA divorcing from Europe? Unthinkable!

…Yet here we are. 

How to Apply It

Your product will succeed? 
Maybe 85%. Not 100%.
Most products?
10-20% max. 

That 15% isn’t pessimism. 
It’s realism. 

It means you plan for failure modes.

Backup suppliers. Pivot options. Cash runway. These exist because nothing is certain.

Like the one time last year I drove into a national park with no rescue kit because “the roads are good”, I got a flat tire. Who would’ve thunk?

The Real Pattern

Overconfidence kills products faster than bad ideas.

The moment you say “this can’t fail,” you’ve stopped preparing for when it does.

The Real Question

What’s your Plan B? If this is 100% certain, you don’t need one.

But it’s not. So you do.

Confidence is good. Certainty is delusion. Know the difference.​​​​​​​​​​​​​​​​

If you aren’t afraid… you will be.

Mental Frameworks for 2026: Falsifiability

If you can’t be proven wrong, you’re not thinking.

That’s why you can’t argue against belief. No thinking. 

The Core Idea

A real thesis has clear conditions that would invalidate it.

If nothing could prove you wrong, you don’t have a hypothesis. 
You have faith.

Why It Matters

Teams build strategies without defining what failure looks like.

They launch products, ignore negative signals, rationalize poor metrics. Because they never said what would disprove their thesis.

Falsifiability forces honest evaluation before you commit.

How to Apply It

Your thesis: this feature will boost retention 15%.

Falsification criteria: if retention increases less than 5% after three months with full rollout, the thesis is wrong.

Now you can’t fool yourself. 
The market will tell you if you’re right. And you’ll have to listen.

The Real Pattern

Unfalsifiable theses never die. 
They just keep demanding more resources. That’s why religion sticks. 

Falsifiable theses fail fast or succeed clearly. Both save time.

The Real Question

What specific outcome would prove this wrong? At what threshold do you pivot?

If your answer is vague or “I’d need to see more data,” you’re protecting ego, not testing ideas.

There has to be a clear failure condition. 

Define failure conditions upfront. Then honor them when they hit.

So much doom, gloom and posturing on LinkedIn now. 
Look, here's something that helps (besides tuning out).

First, gain some perspective.

My feed is full of either WW3 is coming, or AI will kill all jobs, or Climate Change will kill all of us. Maybe some of this is true. Maybe not. Point is:

You think you're the first person to live through difficult times?
People have always lived through hardship scary times.
But they figured out to focus on what they can control.

Is the market going up? Down? 
Is a tech working? Are your skills relevant?
Are you losing your job? Are you getting rejection letters?

It doesn't really matter what the situation is. 
The situation and environment are not in our control.
Only our response to what's going on and our reactions are.

We choose what to make of things.

The best thing we can do is to not be fazed, make our own decisions, then carry on with life as best we can.

At the end of the day, the destination is the same for everyone.

Freischaffender Regisseur

Film- und Musikschaffender aus Berlin!

Mein Spielfilmerstling PLAN B - SCHEISS AUF PLAN A kam 2017 im Verleih der 20th Century Fox in die deutschen Kinos, und ich bin ebenfalls für (Web-)Serien (FOOD-BALLS, 2021, European League of Football), Musikvideos, Werbungen und sonstige Kreativitäten zu begeistern.

Musikalisch bin ich ebenfalls unterwegs: Für Universal Music habe ich das Album PULS - EINS produziert, habe für K.I.Z., Sido, Die Atzen etc. gearbeitet. Meine letzten größeren Veröffentlichungen in dem Bereich waren drei Songs für die Camper (Sido, Knossi, Manny Marc und Sascha): KEINE ANGST, HEILIGENSCHEIN und ANGELCAMP.

taplio

The Battle

ahmet acar

Michael Popescu

Followers

Posts

Engagements

Likes

Performance Scoreahmet acar's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

ahmet acar

Michael Popescu

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}ahmet acar's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance Scoreahmet acar's LinkedIn Performance