Umer Molai

@TryRevvo | Building AI Operating Systems for Biz Owners

[Umer Molai @umeramolai]

Does anyone else feel like using AI is sometimes harder than just doing the task yourself?

I spend more brain power correcting its mistakes than I would've spent just doing the work.

Even as the models get stronger, you still have to be sharp about what you actually delegate to them.

[Umer Molai @umeramolai]

@solopribuilds depends what you're using them for

if it's for the unnecessary repetitive tasks then no, but if you're trying to delegate high-impact tasks to AI then definitely can get exhausting

[Umer Molai @umeramolai]

@danellisona the power of momentum is very real

[Umer Molai @umeramolai]

As I continue to look into different kinds of businesses I notice more & more common themes across all of them:

This is one BIG one:

Most performance metrics are measuring how busy someone looks, not what they're actually getting done.

"Busy" is super easy to count. Impact takes work to define.

Teams track the easy thing and then wonder why they end up rewarding the wrong people.

[Umer Molai @umeramolai]

The fundamentals of running a business doesn't change when implementing AI.

What it does do is expose whether you really had them or not:

- good pricing
- clear ownership
- clean data

AI is what scales all of it, but also scales the mess if that's what you feed it.

Fix the business, then add the leverage.

[Umer Molai @umeramolai]

AI isn’t one skill.

The valuable skills are:

- building / managing agents
- finding distribution
- creating trust
- shipping products
- curating information
- building real communities

Don’t learn AI.

Learn how to create leverage with it.

[Umer Molai @umeramolai]

"AI-first" is backwards.

The companies getting real leverage went process-first, data-first, and AI came third almost as an afterthought.

The ones who went AI-first bought 12 tools, generated a lot of content, and have exactly the same throughput they had last year with a bigger software bill.

The technology is not the constraint. It hasn't been for two years.

The constraint is that most companies don't know what they're doing.

[Umer Molai @umeramolai]

Holyyyy Opus 5 is beating Fable 5 in almost every category & is also half the price..

[Claude @claudeai]

Introducing Claude Opus 5.

It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.

[Umer Molai @umeramolai]

This is one of the most valuable AI datasets available.

The opportunities are endless. Find the industries that aren’t utilizing AI and build solutions for them before anyone else does.

[Claude @claudeai]

You can now ask Claude about the Anthropic Economic Index, our public dataset measuring how AI is used across the economy.

Ask which occupations use AI the most, or what kinds of tasks people are automating, and the answers draw directly from the Index data.

[Umer Molai @umeramolai]

The value of a human didn't go down when AI got good.

It moved.

It used to sit in the doing writing the proposal, pulling the report, answering the email. That's commoditized now, and pretending otherwise is how businesses lose the next three years.

It now sits in three places:

- deciding what's worth doing at all
- being accountable when it goes wrong
- being the person a customer trusts on a bad day

None of those are tasks. That's why they're safe.

The mistake isn't using AI too much. It's using it to stay in the doing layer, faster, instead of climbing out of it.

[Umer Molai @umeramolai]

The repetitive tasks are officially all gone???

You can record yourself showing Claude exactly how you work:

- record your screen & voice explaining each step of a task
- Claude saves exactly what you did & creates a skill that can complete that task with zero effort

No more writing instructions, just show Claude exactly how it’s done.

[Claude @claudeai]

New in Claude Cowork: teach Claude a skill.

Record your screen while you do a task, talk through it as you go, and Claude turns it into a skill it can run again. Find it under Record a skill in the + menu of the Claude desktop app.

Available on Pro, Max, and Team plans.

[Umer Molai @umeramolai]

Three things AI is genuinely great at:

- Reading a lot and telling you what's in it
- Doing a defined task the same way 10,000 times
- Drafting the first version of something a human will fix

Three things it isn't:

- Deciding what matters
- Knowing what you meant
- Caring if it's wrong

Build for the first list. Staff for the second.

[Umer Molai @umeramolai]

saying the space is saturated is just an excuse to not do the work...

the AI Agency space has endless opportunity especially because there's the ability to go super broad with the businesses you're working with early on, rather than only focusing on one niche.

once you crack what niche you like working with best, that's when you double down.

[Umer Molai @umeramolai]

Every company I work with wants an agent.

What they actually need, in this order:

1. One place where the data lives
2. A written version of how the work gets done
3. Someone who owns the process when it breaks
4. Then the agent

Skip 1-3 and the agent becomes a very expensive way to generate confident nonsense from bad inputs.

The unsexy layers are the whole job. The agent is the last 10%.

[Umer Molai @umeramolai]

Delegating your decisions to AI is the fastest way to get worse at making them.

The reasoning IS the skill.

Outsource it and you keep the output but lose the judgment that let you evaluate the output.

Use it to gather. Don't use it to conclude.

[Umer Molai @umeramolai]

the most dangerous thing AI does isn’t replace your job. it invents work that never needed to exist.

reports nobody reads. content nobody asked for. workflows for tasks that didn’t need doing.

that’s not leverage. it’s motion.

point AI at the work that actually moves the needle.

[Umer Molai @umeramolai]

every sunday i spend about an hour with cowork setting up my week

not planning in the vague "set intentions" sense. actual operational prep. here's the routine:

first, it reviews last week

i point it at the client data from the past seven days and ask what moved, what stalled, and what i'd miss if i only skimmed.

it surfaces the stuff buried three layers down that i'd never catch scrolling a dashboard

then it plans the week ahead

every open thread across every client, sorted by what actually needs me vs what's just noise.

by the time i'm done i know exactly where monday goes

then it drafts

the deliverables and content i already know are coming get a first pass on sunday, so monday isn't a blank page. i edit, i don't start from zero

the whole point of this is one thing: nothing slips through the cracks

when you're running ops across multiple clients, things don't blow up loudly. they leak quietly.

a follow-up that never happened. a number that drifted. sunday is where i catch the leak before it becomes a fire

[Umer Molai @umeramolai]

everyone optimizing their prompts. almost nobody optimizing their inputs

i've seen a model "fail" a hundred times in a row, and it wasn't the model

it was garbage data going in. duplicate records. three sources of truth that disagreed

fix the pipes before you touch the prompt

a perfect prompt on broken data is just a confident wrong answer

[Umer Molai @umeramolai]

hot take: most "AI agents" are just expensive scripts wearing a costume

i've watched teams pay per-token for a model to do something a cron job does for free

if the input is the same every time and the steps never change, that's not an agent. that's automation you overpaid for

[Umer Molai @umeramolai]

Everyone tries to automate on day one. That's exactly why it fails.

The order that actually works:

1Context — what the business is
2Connections — live data
3Capabilities — the work it does for you
4Cadence — it runs on a schedule

Identity first. Automation last.