Forem: Max aka Mosheh

This FREE Chrome AI Does Your Web Tasks Like a Human

Max aka Mosheh — Fri, 17 Apr 2026 01:05:19 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Stop doing the same web tasks every day.
You’re not busy.
You’re stuck in browser busywork.

Most people think automation means APIs and weeks of setup.
They’re overthinking it.
The real shift is simpler.

A new type of tool puts a computer-using AI agent inside your browser.
You just tell it what you want done.
It clicks.
It navigates.
It fills forms.
It makes small decisions like a person.

That matters because most “work” lives in tabs.
CRMs.
Portals.
Dashboards.
Vendor sites.
Internal tools.

Here’s a real example.
A teammate records a weekly “lead cleanup” routine once.
It opens the CRM.
Filters duplicates.
Copies fields.
Updates statuses.
Then runs again next week with one click.

↓ Use this simple framework.
↳ Pick one task you repeat 3+ times a week.
↳ Record it once as a routine.
↳ Rerun it on demand.
↳ Schedule it when you’re offline.

⚡ The result is not “AI magic.”
It’s time back.
It’s fewer mistakes.
It’s consistent follow-through.

If an agent can handle your tabs, you can focus on decisions.
What’s the one browser task you’d stop doing forever if you could?

AI “Users” Are Too Nice—Google Built a Bad Bot to Prove It

Max aka Mosheh — Tue, 14 Apr 2026 00:36:32 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Everyone’s talking about AI “users” and simulators.
They’re missing the real risk.
Your AI may pass tests… and still fail customers.

Google built ConvApparel with 4,000+ real shopping chats.
People thought they were chatting normally.
They were secretly routed to a “Good” assistant or a “Bad” one.

Humans got annoyed fast when the bot was awful.
They pushed back.
They left.
They changed what they asked.

But many prompted user simulators stayed calm.
They kept being polite.
Even when the experience was clearly broken.

That gap matters.
Because polite simulators make your product look better than it is.
You ship.
Real customers churn.

The interesting part is this.
Data-trained simulators adapted more like humans.
But a detector still flagged them as fake.
So “more realistic” is not the same as “real.”

Here’s what to do if you test AI systems.
↓
↳ Use real conversation logs whenever you can.
↳ Measure frustration signals, not just task success.
↳ Add “bad assistant” scenarios on purpose.
↳ Track drop-off, re-asks, and sarcasm.
↳ Red-team your sim to be impatient.

If your evaluation never gets messy, it isn’t real.
What’s one moment your users stop being polite?

This Missing Number Predicts If AI Will Take Your Job

Max aka Mosheh — Sat, 11 Apr 2026 00:36:38 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Everyone’s tracking AI exposure like it predicts layoffs.

It doesn’t.

The real predictor is demand.

More specifically: price elasticity.

When AI makes a task cheaper and faster, two things can happen.

Demand explodes.

Or demand barely moves.

If demand explodes, companies often hire more people.

Because there’s suddenly more work worth doing.

If demand barely moves, automation turns into headcount cuts.

Same AI.

Opposite outcome.

Example.

If AI cuts the cost of writing product descriptions by 70%.

An e-commerce brand might go from 5,000 SKUs to 50,000.

That creates new needs: QA, brand voice, compliance, testing, localization.

Demand surged.

Now flip it.

If AI cuts the cost of basic meeting notes by 70%.

Most teams won’t run 10x more meetings.

Demand stays flat.

So the “savings” show up as fewer roles.

Here’s what smart leaders do ↓

↳ Map your work into elastic vs inelastic buckets.

↳ Ask: if this gets 50% cheaper, do we do 2x more?

↳ Invest where volume can scale, not where it can’t.

Economist Alex Imas argues we need a Manhattan Project for elasticity data.

Not just for groceries.

For every major job task we’re betting careers on.

Because guessing your future is not a strategy.

In your industry, where will demand surge when AI drops costs?

He Asked AI to Find a Factory… Cost Dropped $17 $2.50

Max aka Mosheh — Wed, 08 Apr 2026 00:36:27 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Everyone's talking about AI in sourcing.
They're missing the real opportunity.
It’s not speed.
It’s leverage.

A solo online seller asked Alibaba’s AI, Accio, to remake his old best-selling flashlight.
In one chat, it suggested simple tweaks.
Smaller body.
A bit dimmer.
Battery-powered.

Then it surfaced a factory in Ningbo, China.
The estimated unit cost dropped from $17 to about $2.50.
Weeks of back-and-forth became days.

That shift changes the game for small operators.
If your product margin is thin, sourcing speed is survival.
If your volume is low, finding the right factory is the moat.

Here’s the part most people miss.
AI can compress the messy middle.
But it can also create new risk.

↓ A simple guardrail framework before you trust any sourcing AI.

↳ Ask for full supplier transparency.
• Who is the factory.
• Who is the broker.
• What data trained the recommendations.

↳ Lock down security.
• What product files get stored.
• Who can access your specs.

↳ Set clear data guardrails.
• What the tool can retain.
• What must be deleted.

AI won’t replace good judgment.
But it can give you negotiating power faster.

What would you automate first in your sourcing process?

They Pay You $15/hr to Wear an iPhone on Your Head… for Robots

Max aka Mosheh — Sun, 05 Apr 2026 00:36:42 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

A gig worker in Lagos wears an iPhone on their forehead.
They film your chores for $15 an hour.
And the video might end up training a humanoid robot.

Most people hear “robot training data” and think factories.
But a lot of it is laundry.
Dishes.
Ironing for hours.

That’s the hidden shift.
AI isn’t learning only from the internet anymore.
It’s learning from real homes, in real time.

Here’s the part leaders keep skipping.
When you train robots on intimate home footage, you inherit the mess.
Privacy gaps.
Consent gaps.
Safety gaps.

One worker making $15/hour can be life changing locally.
But the downstream value can be massive for the buyer.
The footage becomes a product.
And the worker rarely owns the upside.

If you build, buy, or deploy AI, use this simple checklist ↓
• Who owns the raw footage, forever.
• Who can resell it, and to whom.
• What gets blurred, deleted, or retained.
• How “unsafe habits” are filtered out.
↳ If you can’t answer fast, you have a risk.

Robots will copy whatever we normalize at scale.
That includes the dangerous shortcuts.

If this were your home on camera, what rule would you demand first?

Google Says Quantum Could Crack Crypto in Minutes (Here’s the Safe Warning)

Max aka Mosheh — Thu, 02 Apr 2026 00:36:33 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Everyone’s talking about quantum computing and crypto.
They’re missing the real opportunity.
This is your chance to de-risk your business.

Google Quantum AI shared a new estimate that caught my eye.
They say a future quantum machine could break common wallet math.
Their rough target is under 1,200 logical qubits.
And around 90 million Toffoli gates.
Potentially done in minutes.

The detail that matters most is not the speed.
It’s the signal.
The migration window is always smaller than you think.

They also did something smart.
They used zero-knowledge proofs.
So others can verify the claim without getting an attack recipe.
That’s what responsible disclosure looks like at scale.

If you run a product, a bank, a chain, or any system with long-lived secrets, this is a board-level risk.
Not because quantum is here tomorrow.
Because your data and signatures may need to stay safe for years.

Here’s what smart teams do now ↓
↳ Inventory where public-key crypto lives in your stack.
↳ Flag “harvest now, decrypt later” data.
↳ Add crypto agility so algorithms can swap fast.
↳ Pilot post-quantum options in non-critical paths.
↳ Set a timeline and owner, then rehearse the cutover.

The winners won’t panic.
They’ll migrate before it becomes urgent.

What’s stopping your org from starting a post-quantum plan this quarter?

MIT’s AI Designs Proteins by Their “VIBES” (Motion, Not Shape)

Max aka Mosheh — Mon, 30 Mar 2026 00:41:34 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Most people think protein design is about shape.
They’re overthinking it.
The real advantage is motion.

Proteins aren’t statues.
They’re machines in motion.

MIT built VibeGen, an AI that designs proteins using a vibration “fingerprint,” not just a 3D structure.
That matters because two proteins can look similar but move differently.
And the movement is often what drives function.

VibeGen uses a diffusion generator to propose new designs.
Then it runs an agent loop.
One model designs.
Another model critiques.
They repeat until the motion target is met.

Here’s the wild business lesson.
Different “implementations” can deliver the same “behavior.”
Totally different sequences can share the same motion.

↓ If you build products, teams, or models, steal this framework.

↳ Define the fingerprint, not the artifact.
↳ Optimize for behavior under real conditions.
↳ Add a critique loop that forces iteration.
↳ Reward diversity, as long as outcomes match.

The opportunity isn’t just better proteins.
It’s a new way to design: by what something does over time.

What would you redesign in your work if you optimized for motion, not appearance?

MIT’s Ultrasound Wristband Lets You “Puppet” a Robot Hand

Max aka Mosheh — Fri, 27 Mar 2026 00:41:28 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Most people think robot control needs cameras or sensor gloves.
They’re overthinking it.
A wristband might be the new remote control.

MIT built a wristband that uses ultrasound to watch your wrist tendons move.
No cameras.
No finger sensors.
Just your wrist.

An AI model turns those tendon patterns into real-time finger positions.
It tracks 22 degrees of freedom.
That’s enough detail to copy how you actually move.

Then the wild part.
A robotic hand mirrors your fingers wirelessly.
It can hit piano notes.
It can sink a tiny basketball.

The business lesson is simple.
The winning interface is the one people forget they’re wearing.
When input becomes effortless, adoption explodes.

If you’re building in robotics, VR, or training data, this changes the playbook.
You can capture natural hand motion without a studio.
You can train machines from the wrist up.
You can control devices in places cameras fail.

Here’s a practical way to think about it ↓
• Replace “perfect sensors” with “good enough signals”
↳ Let AI do the cleanup.
• Move computation to the edge
↳ Lower latency wins trust.
• Design for everyday wear
↳ Comfort becomes distribution.

The next big platform shift may not be a headset.
It may be something you put on before you leave home.

Where would you use camera-free hand tracking first?

Robots Don’t Need Stronger Arms—They Need Memory (The Two-Memory Breakthrough)

Max aka Mosheh — Fri, 06 Mar 2026 01:49:32 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Most people think robots fail because the hardware is weak.
They’re overthinking it.
The real problem is memory.

Most robots “forget” what happened seconds ago.
That’s why they miss a grip.
That’s why they panic when something gets hidden from view.

A better approach is simple.
Give robots two kinds of memory.

→ Short-term video memory.
This helps in real time.
It lets the robot correct its grip.
It tracks objects through occlusion.

→ Long-term language memory.
This summarizes what happened.
“Placed three bowls.”
“Added soap.”
“Moved pan to the sink.”

Then you split the brain.
A high-level planner decides the goal.
A low-level controller executes the motion.

Here’s the business lesson.
Systems don’t scale when they can’t remember.
They repeat mistakes.
They can’t recover.
They can’t hand off work.

↓ If you’re building AI products, steal this framework.
↳ Keep a fast memory for immediate corrections.
↳ Keep a slow memory for summaries and handoffs.
↳ Separate planning from execution.

⚡ Result.
Robots can recover from mistakes.
They can follow recipes.
They can clean kitchens.
They can do it in real time on one H100.

What’s one “memory gap” in your product or team that keeps causing repeat failures?

OpenClaw Isn’t a Chatbot—It’s a Hackable Server (Guardrails Most Teams Skip)

Max aka Mosheh — Tue, 03 Mar 2026 01:49:44 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Most people treat agent tools like toys.
That’s how you end up with an exposed server.
And a surprise security incident.

Here’s the truth I learned the hard way.
If a tool spins up a gateway, it is infrastructure.
Infrastructure has blast radius.

The riskiest part is not the model.
It’s what the model can run.
If “skills” are executable code, one plugin can own your machine.
That’s not theory.
That’s how compromises happen.

A real example.
A team shipped an internal agent demo in a day.
They forwarded a port to “test quickly.”
Within 48 hours, scanners found it.
They spent the next week rotating keys and rebuilding a box.
The demo worked.
The trust was damaged.

If you’re building with tools like OpenClaw, use this baseline ↓
↳ Treat the gateway like a production service.
↳ Default deny inbound access.
↳ Use a strong tool model, or actions will misfire.
↳ Run skills in a sandbox.
↳ Never mount your whole home folder.
↳ Lock secrets, scope them tightly, rotate often.
↳ Voice calls are permissioned actions.
↳ Treat them like payments.

Move fast.
Just don’t move fast with root access.

What’s one guardrail you now require before any agent demo ships?

This New AI Agent Actually Remembers (and Uses a Real Terminal)

Max aka Mosheh — Sat, 28 Feb 2026 01:38:18 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Most AI agents feel smart in the moment.
Then you start a new chat.
And it forgets everything you just taught it.

That “amnesia” is why most agent demos never become real workflows.
You keep paying in time, context, and rework.

I noticed something important this week.
The next wave of agents won’t win on prompts.
They’ll win on memory and execution.

Nous Research just shipped Hermes Agent, and the idea is simple.
It turns your wins into searchable markdown Skill Documents.
So the next time you need that workflow, it repeats it quickly.

The second shift is even bigger.
Persistent terminal access.
Local.
Docker.
SSH.
Even HPC containers.

That means the agent can actually run.
Not just talk.
It can work for hours while you sleep.
Then ping you when it’s done.

Imagine a weekly ops task that takes 90 minutes.
If an agent learns it once and reruns it every week, that’s 78 hours saved per year.
Per person.

Here’s the playbook ↓
↳ Pick one repeatable workflow you hate.
↳ Define “done” in plain English.
↳ Store the steps as a reusable skill.
↳ Give it a real terminal and a safe sandbox.
↳ Review outputs like you would a junior teammate.

The hidden advantage is compounding.
Your best process becomes a repeatable asset.

What’s the one workflow you’d most want an agent to learn once and run forever?

Google’s New AI Turns Photos Into Songs (And You Can Steer It Live)

Max aka Mosheh — Sun, 22 Feb 2026 02:01:18 +0000

Join our FREE AI Community: https://www.skool.com/ai-with-apex/about

Everyone’s talking about Lyria 3 making music from photos.
They’re missing the real opportunity.
Creation just became live, not linear.

DeepMind’s new model can take an image and a vibe prompt.
Then it generates a ~30-second song with vocals and instruments.
Not after you hit “render.”
While you’re still steering.

The part I noticed most is the streaming.
Audio comes in 2-second chunks.
That means you can guide the track mid-flight.
Like directing a session, not exporting a file.

This changes how teams work.
Marketing can prototype sound for a campaign in minutes.
Product teams can test sonic branding faster.
Creators can iterate without booking studio time.

A simple example.
You’re launching a winter collection.
You upload a snowy street photo.
You prompt “warm, nostalgic, lo-fi pop.”
In under a minute, you have a draft to react to.
Then you adjust mood, tempo, and energy live.

Here’s the framework I’d use this week ↓
↳ Start with one image that already performs.
↳ Write 3 vibe prompts, not 30.
↳ Generate 5 drafts, then pick one direction.
↳ Iterate live until it fits the brand.
↳ Document what changed and why.

One more detail matters.
SynthID watermarks the waveform for traceability.
That’s not a footnote.
That’s how this becomes usable at scale.

Where would “live-steerable” music help you most?