DecentralizedTech and digitalization | Decentralizednews and updates | Ex-Amazon ops | AI innovation and education | Free Newsletter
California, USA
Joined October 2022
- Tweets3.2K
- Following171
- Followers136
- Likes7.4K
DecentralizedHQ retweeted
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.
It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
anthropic.com/glasswing
DecentralizedHQ retweeted
Computer now connects with Plaid to link bank accounts, credit cards, and loans.
Track spending in detail, build custom budget tools, and visualize your net worth alongside your investment portfolio.
MIT researchers found that 83% of people who used ChatGPT to write essays couldn't quote their own work afterward.
They called it "cognitive debt."
I read that and immediately thought of every business I've seen run the same way.
Not with AI — but with people.
When a process lives entirely in one person's head, the business is doing the same thing those essay writers did — producing results without owning the understanding behind them.
And just like those participants, the moment the crutch is gone, so is the capability.
I rebuilt forgotten SOPs in a large fulfillment operation.
Trained people on them. Posted them where the work happened.
Watched the floor run cleaner when they were followed.
Then I left.
And the scramble to rebuild what I'd been quietly carrying began.
That's a story about what organizational cognitive debt costs — and how to fix it before it costs you.
This week's post breaks down:
→ Why the knowledge in your head is your biggest liability
→ The 3 SOPs every small business needs to build first
→ How to draft one in under 30 minutes using AI
→ A free template you can use today
The SOP Problem: Why Most Small Businesses Can’t Scale, by @DukeM_ai open.substack.com/pub/dukemu…
DecentralizedHQ retweeted
A lot of organizations and small businesses are doing the same thing right now.
They find a bottleneck. Something is slow, something is costing money, something is not working the way it should. And instead of stopping to figure out why, they go looking for a tool to fix it.
AI is the tool most of them are landing on.
I get it. The promise is real. The case studies are everywhere. And when you are running a small operation with limited time and limited margin for error, the idea that a tool can solve the problem feels like a lifeline.
But here is what I watched happen on the warehouse floor at Amazon, and what I keep seeing play out at every scale:
The tool does not fix the system. It inherits the system.
If your process has a measurement problem, the AI will optimize for the wrong measurement, faster and more consistently than any person would.
If your team has been moving too fast to catch errors, the AI will move faster and miss more.
If nobody had clear ownership of quality before the tool arrived, nobody will after it either.
Amazon ran this at global scale and produced documented failures across multiple AI investments. McDonald's ran the same experiment in one of the simplest operational environments imaginable (a drive-thru) and got the same outcome.
The environment changed. The mechanism did not.
Most small businesses will never make the news when this happens. The failure just looks like a tool that did not work out, a few wasted months, a budget that got absorbed with nothing to show for it.
The question worth asking before any AI decision is not which tool, but whether the workflow underneath it is actually ready to be automated.
Fix the system first. Then decide what gets automated.
That is what this week's post on is about.
When the Tool Inherits the Problem: Why AI Adoption Without Diagnosis Scales What Is Broken, by @DukeM_ai open.substack.com/pub/dukemu…
A lot of organizations and small businesses are doing the same thing right now.
They find a bottleneck. Something is slow, something is costing money, something is not working the way it should. And instead of stopping to figure out why, they go looking for a tool to fix it.
AI is the tool most of them are landing on.
I get it. The promise is real. The case studies are everywhere. And when you are running a small operation with limited time and limited margin for error, the idea that a tool can solve the problem feels like a lifeline.
But here is what I watched happen on the warehouse floor at Amazon, and what I keep seeing play out at every scale:
The tool does not fix the system. It inherits the system.
If your process has a measurement problem, the AI will optimize for the wrong measurement, faster and more consistently than any person would.
If your team has been moving too fast to catch errors, the AI will move faster and miss more.
If nobody had clear ownership of quality before the tool arrived, nobody will after it either.
Amazon ran this at global scale and produced documented failures across multiple AI investments. McDonald's ran the same experiment in one of the simplest operational environments imaginable (a drive-thru) and got the same outcome.
The environment changed. The mechanism did not.
Most small businesses will never make the news when this happens. The failure just looks like a tool that did not work out, a few wasted months, a budget that got absorbed with nothing to show for it.
The question worth asking before any AI decision is not which tool, but whether the workflow underneath it is actually ready to be automated.
Fix the system first. Then decide what gets automated.
That is what this week's post on is about.
When the Tool Inherits the Problem: Why AI Adoption Without Diagnosis Scales What Is Broken, by @DukeM_ai open.substack.com/pub/dukemu…
DecentralizedHQ retweeted
Everyone's talking about "AI agents." But most people are using the term to mean two very different things.
Here's the distinction that actually matters:
An AI agent is a structural thing. It's standalone software built for one specific task — narrow, rule-based, low autonomy. A spam filter, a scripted chatbot, a fraud detector. You know the inputs. You know the expected outputs, it doesn't improvise.
Agentic AI is a capability thing. It's an operational mode — emergent, adaptive, multi-step. It plans, reasons, and uses tools to get there. Claude Code, AutoGPT, Devin. The task is open-ended, and the system figures out the path as it goes.
Same industry. Very different risk profiles.
Why does this matter for small businesses?
Because most AI failures aren't from bad tools — they're from mismatched expectations. Someone deploys an "AI agent" expecting agentic behavior. Or they hand an agentic system a task it should never be trusted to self-direct.
The fix isn't more AI literacy in the abstract, it's knowing which type of system you're dealing with before you automate anything.
Fix First. Automate Second.
What's one place you've seen "AI agent" used when someone clearly meant something else? Comment your thoughts below and lets have a discussion.
Everyone's talking about "AI agents." But most people are using the term to mean two very different things.
Here's the distinction that actually matters:
An AI agent is a structural thing. It's standalone software built for one specific task — narrow, rule-based, low autonomy. A spam filter, a scripted chatbot, a fraud detector. You know the inputs. You know the expected outputs, it doesn't improvise.
Agentic AI is a capability thing. It's an operational mode — emergent, adaptive, multi-step. It plans, reasons, and uses tools to get there. Claude Code, AutoGPT, Devin. The task is open-ended, and the system figures out the path as it goes.
Same industry. Very different risk profiles.
Why does this matter for small businesses?
Because most AI failures aren't from bad tools — they're from mismatched expectations. Someone deploys an "AI agent" expecting agentic behavior. Or they hand an agentic system a task it should never be trusted to self-direct.
The fix isn't more AI literacy in the abstract, it's knowing which type of system you're dealing with before you automate anything.
Fix First. Automate Second.
What's one place you've seen "AI agent" used when someone clearly meant something else? Comment your thoughts below and lets have a discussion.
DecentralizedHQ retweeted
"Vibe coding" is the hottest term in AI right now — and most people are either overselling it or completely missing the risk.
Let me break it down honestly.
━━━━━━━━━━━━━━━━━━━━
WHAT IT ACTUALLY IS
━━━━━━━━━━━━━━━━━━━━
Vibe coding (coined by AI researcher Andrej Karpathy) is the practice of describing what you want in plain language, letting AI write the code, accepting the output without fully understanding it, and iterating by prompting — not by engineering.
When something breaks, you describe the error back to the AI and let it fix it.
You're not a developer. You're a director.
That shift has more implications than most people realize.
━━━━━━━━━━━━━━━━━━━━
THE REAL UPSIDE
━━━━━━━━━━━━━━━━━━━━
Let's be fair — vibe coding has genuine value:
✅ Dramatically lowers the barrier to building tools
✅ Prototype in hours, not weeks
✅ Great for internal, low-stakes MVPs
✅ Lets you focus on WHAT to build, not HOW
✅ Gives non-technical founders actual leverage
For small business owners who couldn't afford a developer before? This is real.
But the keyword is: when used right.
━━━━━━━━━━━━━━━━━━━━
THE HIDDEN COST
━━━━━━━━━━━━━━━━━━━━
Here's what the hype skips over:
❌ Code you don't understand becomes debt you can't manage
❌ Bugs hide in systems you never audited
❌ Security gaps you won't catch without the right context
❌ When it breaks in an unexpected way — you may have no path forward except starting over
This isn't just a coding problem. It's a workflow problem.
━━━━━━━━━━━━━━━━━━━━
I'VE SEEN THIS PATTERN BEFORE
━━━━━━━━━━━━━━━━━━━━
Not in code. On warehouse floors.
Rush a new system live. Skip the process audit. Celebrate the speed.
Then watch the errors stack up — damaged inventory, overwhelmed staff, broken SOPs nobody documented, and a floor that's moving fast in the wrong direction.
Vibe coding is the software version of that same mistake. The output looks functional. The underlying structure is fragile.
━━━━━━━━━━━━━━━━━━━━
THE CORE PROBLEM
━━━━━━━━━━━━━━━━━━━━
Vibe coding is "automate first, understand later."
But automation applied to a broken or unexamined workflow doesn't fix the problem — it accelerates it.
Speed without structure isn't efficiency. It's deferred chaos with a better UI.
━━━━━━━━━━━━━━━━━━━━
SO WHEN SHOULD YOU USE IT?
━━━━━━━━━━━━━━━━━━━━
Vibe coding IS worth using — selectively:
✅ Internal tools you'll maintain yourself
✅ Quick prototypes to test an idea before committing
✅ Personal projects with low stakes and no clients on the line
🚫 Client-facing systems you can't fully explain
🚫 Automations you can't troubleshoot when they fail
🚫 Anything that touches sensitive data or real money
━━━━━━━━━━━━━━━━━━━━
THE REAL QUESTION
━━━━━━━━━━━━━━━━━━━━
Stop asking: "Can AI build this for me?"
Start asking: "Do I understand what I'm deploying — and can I stand behind it when something goes wrong?"
Vibe coding is a tool. Not a strategy.
AI is not the strategy. The workflow is.
Fix First. Automate Second.
♻️ Repost if this reframed how you think about AI tools. Follow for more on building AI into your business without the hidden costs.
"Vibe coding" is the hottest term in AI right now — and most people are either overselling it or completely missing the risk.
Let me break it down honestly.
━━━━━━━━━━━━━━━━━━━━
WHAT IT ACTUALLY IS
━━━━━━━━━━━━━━━━━━━━
Vibe coding (coined by AI researcher Andrej Karpathy) is the practice of describing what you want in plain language, letting AI write the code, accepting the output without fully understanding it, and iterating by prompting — not by engineering.
When something breaks, you describe the error back to the AI and let it fix it.
You're not a developer. You're a director.
That shift has more implications than most people realize.
━━━━━━━━━━━━━━━━━━━━
THE REAL UPSIDE
━━━━━━━━━━━━━━━━━━━━
Let's be fair — vibe coding has genuine value:
✅ Dramatically lowers the barrier to building tools
✅ Prototype in hours, not weeks
✅ Great for internal, low-stakes MVPs
✅ Lets you focus on WHAT to build, not HOW
✅ Gives non-technical founders actual leverage
For small business owners who couldn't afford a developer before? This is real.
But the keyword is: when used right.
━━━━━━━━━━━━━━━━━━━━
THE HIDDEN COST
━━━━━━━━━━━━━━━━━━━━
Here's what the hype skips over:
❌ Code you don't understand becomes debt you can't manage
❌ Bugs hide in systems you never audited
❌ Security gaps you won't catch without the right context
❌ When it breaks in an unexpected way — you may have no path forward except starting over
This isn't just a coding problem. It's a workflow problem.
━━━━━━━━━━━━━━━━━━━━
I'VE SEEN THIS PATTERN BEFORE
━━━━━━━━━━━━━━━━━━━━
Not in code. On warehouse floors.
Rush a new system live. Skip the process audit. Celebrate the speed.
Then watch the errors stack up — damaged inventory, overwhelmed staff, broken SOPs nobody documented, and a floor that's moving fast in the wrong direction.
Vibe coding is the software version of that same mistake. The output looks functional. The underlying structure is fragile.
━━━━━━━━━━━━━━━━━━━━
THE CORE PROBLEM
━━━━━━━━━━━━━━━━━━━━
Vibe coding is "automate first, understand later."
But automation applied to a broken or unexamined workflow doesn't fix the problem — it accelerates it.
Speed without structure isn't efficiency. It's deferred chaos with a better UI.
━━━━━━━━━━━━━━━━━━━━
SO WHEN SHOULD YOU USE IT?
━━━━━━━━━━━━━━━━━━━━
Vibe coding IS worth using — selectively:
✅ Internal tools you'll maintain yourself
✅ Quick prototypes to test an idea before committing
✅ Personal projects with low stakes and no clients on the line
🚫 Client-facing systems you can't fully explain
🚫 Automations you can't troubleshoot when they fail
🚫 Anything that touches sensitive data or real money
━━━━━━━━━━━━━━━━━━━━
THE REAL QUESTION
━━━━━━━━━━━━━━━━━━━━
Stop asking: "Can AI build this for me?"
Start asking: "Do I understand what I'm deploying — and can I stand behind it when something goes wrong?"
Vibe coding is a tool. Not a strategy.
AI is not the strategy. The workflow is.
Fix First. Automate Second.
♻️ Repost if this reframed how you think about AI tools. Follow for more on building AI into your business without the hidden costs.
Nate B. Jones said good taste and specification are the skills that survive AI.
He's right. But he left something out.
What happens when your specifications are wrong — and nobody catches it?
I watched this play out in Amazon fulfillment. Leadership pushes a productivity program. Targets go up. Training gets compressed. The dashboard looks green — but everyone on the floor knows the process is breaking underneath them.
Now replace compressed training with AI-generated SOPs nobody has bandwidth to verify.
The output looks finished. Sounds authoritative. Passes the surface review.
Then it runs at scale — with the error baked in.
That's the quality decay risk.
When production cost collapses, verification doesn't scale with it. Human judgment is still slow. Still expensive. And if you haven't built checkpoints to interrogate AI outputs before they become the system — you're not moving faster. You're failing faster.
Good taste at the front end is real.
But you also need structure on the back end that treats AI output as provisional — not finished — until it's validated against reality.
AI doesn't fix a broken process. It scales it.
Newsletter in bio 👆 — breaking this down piece by piece.
DecentralizedHQ retweeted
AI doesn't decay quality. Organizations do.
The risk that is often overlooked in the rush to automate everything is real and present. As we embrace automation, it's crucial to recognize that the effectiveness of AI is directly tied to the structures and processes within organizations. Quality can only be maintained if organizations prioritize the right practices and frameworks to support AI initiatives.
Let's focus on ensuring that our organizational strategies align with the capabilities of AI to truly harness its potential. I take a deep dive in my latest article below.
AI Doesn't Decay Quality. Organizations Do., by @DukeM_ai open.substack.com/pub/dukemu…
AI doesn't decay quality. Organizations do.
The risk that is often overlooked in the rush to automate everything is real and present. As we embrace automation, it's crucial to recognize that the effectiveness of AI is directly tied to the structures and processes within organizations. Quality can only be maintained if organizations prioritize the right practices and frameworks to support AI initiatives.
Let's focus on ensuring that our organizational strategies align with the capabilities of AI to truly harness its potential. I take a deep dive in my latest article below.
AI Doesn't Decay Quality. Organizations Do., by @DukeM_ai open.substack.com/pub/dukemu…
DecentralizedHQ retweeted
How to create interactive charts inside Claude (no code):
Step 1: Download Claude (claude. com/ download)
Start your prompt with "Create an interactive chart about [...]."
Add this to make sure it makes it interactive within the chat, and not code: "Don't code. Interactive chart only."
Step 2: Be specific with your prompt
Include variables and scenarios.
Ask for pessimistic/optimistic views.
Ask for interactions, not static data.
Prompt examples in my article below.
Step 3: Combine with web search
Ask Claude to search the web first.
Find consensus on real data.
Then generate the chart from findings.
Step 4: Keep prompting for depth
Don't stop at the first chart.
Ask follow-ups for more tactics, scenarios, data.
Claude keeps going. I couldn't find the end.
Step 5: Share with your team
Click 'Share' top right.
Select 'workspace.'
Copy the link directly.
Works with Claude team accounts.
Step 6: Share with a client
Follow up with: "I need to send it to a client. Code it."
Claude codes a full interactive website.
Download the HTML file.
Upload to Google Drive.
Send the Drive link.
I worked on Amazon's floor.
I watched the same pattern play out that just cost them 6.3 million orders.
Different building. Different tools. Same cultural problem.
Amazon mandated that 80% of engineers use its AI tool Kiro weekly — and tracked it as a corporate OKR.
Not whether the tool improved outcomes.
Not whether oversight could keep up.
Just: are you using it enough?
Then Kiro deleted a production environment.
Then Q helped trigger an outage — 120,000 lost orders.
Then a second outage wiped 99% of orders for an entire day.
6.3 million sales. Gone. In three days.
Amazon's official response?
"User errors. Not AI errors."
I've heard that before.
Just in different language, on a different shift, about a different kind of failure.
The pattern I kept seeing on the floor:
↳ Speed gets treated as strategy
↳ Volume gets treated as value
↳ The people raising quality concerns get treated as friction
↳ When it breaks — blame the workers, not the system design
Amazon didn't have an AI problem.
They had a leadership culture problem that AI made significantly more expensive.
And that pattern doesn't stay in large corporations.
It scales down into every business that mistakes activity for progress.
—
Are you measuring what AI is doing for your business?
Or just whether you're using it?
That question is the difference between a tool and a liability.
DecentralizedHQ retweeted
This is wild.
143 million people thought they were catching Pokémon. They were actually building one of the largest real-world visual datasets in AI history.
Niantic just disclosed that photos and AR scans collected through Pokémon Go have produced a dataset of over 30 billion real-world images. The company is now using that data to power visual navigation AI for delivery robots.
Players didn't just walk around with their phones. They scanned landmarks, storefronts, parks, and sidewalks from every angle, at every time of day, in lighting and weather conditions that staged photography would never capture. They documented the physical world at a scale no mapping company with a fleet of vehicles could have replicated on the same timeline or budget.
Niantic collected this systematically, data point by data point, across eight years, while users thought the only thing at stake was catching a rare Charizard.
The most valuable AI training datasets in the world aren't being assembled in data centers. They're being built by people who have no idea they're building them.
POKÉMON GO PLAYERS TRAINED 30 BILLION IMAGE AI MAP
Niantic says photos and scans collected through Pokémon Go and its AR apps have produced a massive dataset of more than 30 billion real-world images.
The company is now using that data to power visual navigation for delivery robots, letting them identify exact locations on city streets without relying on GPS.
Source: NewsForce
DecentralizedHQ retweeted
When your mind is flooded with noise, return to knowledge.
Books slow your thinking and deepen your focus.
They sharpen judgment and expand your perspective.
While the world scrolls endlessly, the reader grows quietly stronger.
Feed your mind with ideas, not distractions.
DecentralizedHQ retweeted
There are 5 workflow gaps quietly killing growth. And most owners don't see them until something accelerates them into the open.
The real problem underneath all five?
Not a shortage of tools. Not a shortage of talent.
A shortage of people willing to protect the right way of doing things — even when the fast way is right there.
Before you add AI to your business, ask yourself which of these gaps is running underneath it.
Because AI doesn't fix gaps. It accelerates whatever is already there.
Why AI Won't Save a Broken Workflow, by @DukeM_ai open.substack.com/pub/dukemu…