THIS is Why You're Still Slow Even With AI (The Bottleneck Moved--Here's What to Do About It)

The one constant right now is chaos. I

hear it over and over again from folks.

The rate of change, the sheer

unpredictable chaos of AI is very

difficult to tell what's up and what's

down. So, in this video, I want to

simplify it. I want to zero in on some

of the underlying drivers that are

shifting what truly AI native working

looks like. Why most of our work habits

are now optimizing for the incorrect

thing. And I want to give you some

specific habits, eight of them in fact,

that you can break that help you start

working in a more AI native way. But

first, I want to start with two scenes

from the last month. Number one,

enthropic shipping co-work a full

product feature. It was built in 10 days

with just four people. It was written

entirely in claude code. And claude

code, mind you, is an entire product

that is less than a year old. And so

these folks have not had years working

AI natively to do this. The anthropic

team is evolving as they go. Meanwhile,

scene two at your company and many

companies I've worked at in the past,

there's a conference room where I

guarantee you a leader is asking for a

30-day implementation roadmap or a

three-month implementation roadmap for

their AI strategy with phases and

milestones and resource allocation and a

plan to protect capacity. Listen,

execution capacity isn't scarce anymore.

10 days, four people, and they're

shipping 60 to 100 releases daily.

Execution capacity is not the problem.

When we build our AI strategies, we're

frequently asking for help or asking for

guidance on the thing that is no longer

scarce and no longer requires

efficiency. We have spent so much of our

business lives assuming that execution

capacity is scarce. Anthropic and many

other organizations now are showing that

it's not scarce anymore. And yet every

organization, all of our individual work

habits, anyone who's had a career more

than three, four years in the industry,

it's all built around the implicit

answer to one question. What's expensive

here? And the answer has been execution.

For most of our careers, building things

required scarce hours from scarce people

with scarce skills. Finding really good

engineers was really hard. Training them

took a long time, maybe years. Every

hour of their time was precious. And so

we evolved elaborate rituals to protect

that capacity. Planning processes,

approval gates, specs, PRDs, meetings to

align before anybody built. All of it

was designed to protect the precious

engineering execution time so it

wouldn't be wasted on the wrong

problems. And that made sense. When the

meeting to discuss a feature takes much

less time than the time to build it, you

definitely want to hold the meeting so

you get it right. When gathering the

requirements costs much less time than

writing the code, you want to gather the

requirements first. When rework is

really expensive, you want to plan

really meticulously so you avoid rework.

Now I know and you know that there have

been movements in software to make this

easier. Agile comes to mind. But even

then, engineering work remained

expensive and agile was a response to

the idea that you had to optimize

engineering work over time to deliver

value. AI has inverted the entire cost

ratio and changed the way we think.

We're no longer in a world where we

argue about waterfall versus agile.

We're in a world where we're talking

about a different kind of work entirely.

Agile never imagined a world where

everybody commits code in the

organization. Curser, the AI code

editor, they went from a million dollars

to $500 million in annual revenue faster

than any SAS company in history. And

they're not done yet. They are launching

Cursor for designers. And they're not

launching it as a separate product.

They're just relentlessly shipping

features inside Cursor that make it

easier and easier and easier for other

job families to use that tool. What used

to be an impossible product expansion is

now another day at cursor. What used to

be a new feature like claude co-work

that took months to plan and ship is now

10 days for four people. Coinbase

engineers report that single people are

now refactoring, upgrading or building

new code bases in days. Same story. So

in that world, the meeting to discuss a

feature now takes longer than building

the feature. The PRD can take longer

than the prototype. The planning process

can take longer than shipping three

version and seeing which ones work. The

bottleneck has moved, but our work

habits are still stuck in the way we've

worked most of our careers. There's a

manufacturing principle that explains a

lot of what's happening here. When you

eliminate a bottleneck in a

manufacturing system, the bottleneck

doesn't actually disappear. It moves

somewhere else downstream in the system.

you see the new crunch point in the

system. Resistance is never destroyed,

it's relocated. And so when you gain

efficiency in one place, you see new

constraints. That's what's happening

now. And so for decades, the constraint

in knowledge work was execution. Now

that AI has largely removed that

constraint, the bottleneck has shifted.

It hasn't disappeared. A lot of people

will tell you it's disappeared. It's not

disappeared. is shifted to clarity, to

ambition, to distribution, and to

relationships. And we're still running

around using work habits that are

designed to protect execution capacity.

We are optimizing for the old work

constraint while the new one is

compounding. So, let's explore where

that bottleneck went. Clarity. Do you

actually know what's worth building?

That is now a billion-dollar question.

You can now build faster than you can

think. Every day now I see new startups

come out of stealth that claim they can

build a business with a prompt. In 2026

you are going to see people try that and

do that and some of them will make

money. The ones that make money are the

ones that will have clarity. The ones

that know what's worth building. Because

it turns out the bottleneck was never

putting the product on the website. It's

knowing what product the customer wants.

PRDs were always a substitute for

clarity. They were a big hedge against

expensive rework. They were a way to

disambiguate and get to some clarity

when you were facing a potentially risky

investment, a six figure, seven figure

investment in engineering time to get to

a prototype. But now that can cost more.

Writing a PRD can cost more than

shipping the whole thing. And I'm not

kidding. I have seen PRD cycles in my

career at big companies take longer than

Claude took to ship all of co-work. The

second bottleneck is around ambition.

Are you swinging hard enough when

shipping requires

a quarter of engineering time? Then

small bets can make sense because I mean

that's agile style, right? You don't

want to you don't want to lock up

engineering and increase your risk. You

might have three or four shots per year.

If you can increase that a little bit

with agile style thinking, then that's

great. But what if that's no longer a

constraint? What if you can take, you

know, every 10 days you can ship, every

week you can ship, it's 50 swings a

year. Suddenly your risk is timidity.

Your risk is lack of courage. The danger

isn't necessarily building the wrong

thing because you got 50 shots to build

the right thing. The danger is not

building enough things toward a larger

vision that is really transformative for

the customer. And so I think that we're

going to see a lot of cases where people

are using AI to build horseless

carriages, which is the old name for

cars. It's what we called cars because

we didn't have a mental model for a car

when we first got them. We called them

horseless carriages. We will see a lot

of products that are horseless carriages

floating around. And what we need to do

is have the ambition and the eyes to see

what is that 10x better product in our

particular domains and shoot for that

with multiple releases as quickly as we

can. Distribution is another bottleneck

that's becoming clear right now. When

everybody can build, product is not

really the moat that it was. Getting it

into people's hands is a mode. My

favorite example right now is Cognition

makers of the AI coding agent Devon

chose not to go and get distribution

themselves even though they have a

product that is an agentic coder that

seems very on trend for 2026. Instead,

they partnered with Infosys and they're

deploying Devon across Infosys's third

of a million person team and hitting all

of their global client base. Why?

Because Infosys has distribution.

Infosys has decades of enterprise

relationships. The technology is the

easy part. Reaching customers is hard.

And Cognition realized that an

established brand was a way to do that.

And that brings me to the last

bottleneck, relationships. When

capabilities are compounding really

quickly and platforms and channels keep

shifting and what worked last quarter

might not be the right way to do it next

quarter, the thing that is durable is

relationships. You can't vibe code a

relationship. And this is going to be a

fractal truth. By which I mean it's true

for individuals and it's true for

companies. Companies in business

relationships are going to have durable

advantages by investing in those

relationships. And you in your career

individually will have a durable

advantage by investing in your

professional relationship so that you

are known as a trusted person to do work

with because if you have to turn to

someone and technical skills are rapidly

becoming a commodity you're going to

turn to someone that you can trust to

deliver on your work and that's a

relationship thing. So, we've taken a

tour through. We've seen how some of

these constraints are shifting. We've

seen how the AI native ways of working

are upending our assumptions about

execution. How do our habits need to

change? Right now, most of the work

habits we've embied are riskmanagement

rituals designed for a world where

execution was expensive. And they've

sort of calcified. They've they've

clotted into defaults that persist even

though all of the unit economics in our

world have flipped. You've probably felt

this. You probably felt a sense that

you're spending more time prepping for

the work than you probably should. That

there should be an easier way to do

this. That maybe you're protecting

something that doesn't need as much

protection anymore. Maybe you're

preparing for a meeting that doesn't

need the kind of ritual it needs

anymore. Let me get specific with you

then. Let me suggest to you habits that

made sense in the old model and are

probably now actively costing you so

that you can flip them and they can

become the seeds of a more AI native way

of working. Number one is the permission

loop. The old logic is that doing

something was typically expensive in

time. So check before you do. Get buy

in. Make sure you're building the right

thing before you spend precious

resources. Now we had management books

for a long time that said we need to

have more autonomy. we need to push

delegation down, but those still didn't

change the fundamental understanding

that execution was costly. And so, yeah,

you could take bias for action and write

up a proposal, but if you're going to

spend a quarter of engineering time,

someone's going to need to sign off on

that. That logic is now broken. Asking

even for relatively large things still

now takes longer than doing the email

thread to get approval can take more

time than building the prototype. Now

it's that's the world we live in. The

Slack conversation to confirm direction

can take longer than trying both

directions and just seeing what works.

We are in a world where Manis just

launched a feature that literally builds

the presentation that you talk about in

the meeting as you're having the

meeting. approval processes were

designed to reduce risk, but they don't

anymore. They just add slowness. And so

this means that our organizations need

to shift because people need to be free

to default to doing, to building the

rough version, to showing it, to asking

forgiveness when needed. we're going to

make mistakes and to committing to a

broader vision, a broader direction,

something in line with the ambition of

the business that they can reliably ship

against autonomously. We need leaders

who are able to cast that wider vision

so teams can be more independent and

ship relentlessly and autonomously

against that vision. That's how you

break the permission loop. Habit number

two, polish as procrastination. The old

logic was you get one shot so make it

count. If execution is expensive, please

don't waste it on something halfbaked.

Look, I saw this show up in the when I

was writing product requirement docs and

PR FAQs at Amazon. You had one shot with

a particular person you wanted to meet

with and do a review with. You had to

make it count. You had to polish it. You

had to go through different reviews.

Quality really mattered. And we spent a

lot of time polishing to make sure those

PRFAQs were as good as they could get.

that's now broken. People are spending

80% of their time on the last 20% of

quality

when the marginal value of that polish

is dropping quickly. And I want to be

careful here. I am not saying that good

thinking is going out of style. I am

saying that polish is becoming a way to

avoid getting your ideas rapidly in

touch with reality. Because before when

you needed to polish something, yeah,

maybe you overpolished. It was still

possible to do that, but at the end of

the day, what you were trying to do is

sharpen your ideas so you could derisk

execution. And that made sense. Now,

even a directionally correct, ambitious

idea is enough to get going, which means

that the value of shipping ugly is up

and the rough version that exists is

going to beat the polished version that

doesn't. and you can always improve on

it. And by the way, I'm not saying that

consumers won't value polish. I think

there's a huge amount of money to be

made in polished UI for AI products. I

think we have a lot of rough edges on a

lot of our AI products. I use them a ton

and so I don't mind them, but I have no

illusions about the fact that we're

leaving billions of dollars in market

value on the table by not fully

polishing out our products because we're

shipping so fast. So take that as a

goal, something we need to iterate on.

But if you're getting started, if you're

trying to get there quickly, you just

got to ship and then you got to

relentlessly optimize from there. And

then you can polish as quickly as you

can. Which by the way, if you're

shipping fast, that path to polish looks

really nice. I think Notebook LM is a

great example here. They shipped, they

got into market, they saw reaction, and

they've been polishing that UI ever

since. Habit number three, meetings as a

default. Look, meetings are a function

of people and one of the consequences of

AI native teams is they tend to be

smaller and so you get less meetings.

But regardless of the size, the old

logic was that you would get alignment

before you got action. You would get

everybody in the room so we don't waste

expensive execution time building. You

want to talk it through that now takes

longer than building the prototype. An

hour of six people's time is 6 hours of

work. And that's often enough to just

build the thing. Meetings still feel

responsible, don't they? Meetings still

distribute accountability. If the

meeting decided something and it was

wrong, it's not our fault. The meeting

decided. And the worst part is this.

Meetings about what to build often don't

resolve what to build. They don't even

answer the question. They'll surface

opinions. They'll create action items

and they'll create delays. AI is

relentless about this. AI will show the

truth here. Just build it. You have the

time. Build it and show it. It might be

bad. It might be good, but just build

it. You can replace the meeting with a

product demo. In fact, maybe the next

time you schedule a meeting, just ask

yourself, what if I built the rough

version of this and showed people

instead? That's what a lot of these AI

native companies are doing. In fact,

that's one of the foundational build

principles that is animating the culture

of cursor right now. Cursor thinks a lot

about how code is a way of getting your

your ideas into contact with reality and

software. And one of the things that

they've talked about in interviews over

and over again is that they are about

taking some of the croft, some of the

extra, some of the planning that

protected execution out of work so that

everybody can just get in and like

designers can commit and product can

commit and engineering can commit and

you can see your ideas meet the code and

that saves meetings. Meetings are no

longer a default. Number four, I call

this structured waiting. I there may be

a better word. The old logic was that

coordination was important. Waiting for

feedback was important because again

you're wanting to protect execution time

and keep it aligned. Everyone's time is

precious. So you must respect the

process. Wait for feedback. Wait for the

sync. Wait for someone to unblock you.

So much of work in older larger

corporations is waiting. Waiting for

approval. Waiting for feedback. Waiting

for the next meeting for someone else to

do their part. But most of what you're

waiting for doesn't need to be waited

for. and you end up outsourcing your

momentum to other people's calendars.

So, I'm going to suggest that you just

stop waiting. Do the next thing while

you wait for feedback on the first

thing. Assume the answer is yes. This is

another example of where leaders need to

set the tone here because leaders need

to cast a vision large enough that their

teams always have something to work

against, do not have to get stuck

waiting and where they know that there's

space to get forgiveness later when

they're working in the correct

direction. And we have tried to drill

this in, if you're a leader, you've

often tried to drill this in as a habit

even before AI came along. But the cost

here is much larger because waiting an

hour in the 2010s was waiting an hour.

Waiting an hour now is waiting a

prototype. Like it's just much more

valuable. And so if you're blocked on a

decision, make a provisional decision

now. Let people know what you picked and

keep moving and see what happens. But

get rid of structured waiting. Number

five, we've inverted planning and doing.

The old logic was very much measure

twice, cut once. Planning is cheap.

Execution is expensive. I actually had

that drilled into me when I was coming

up in product. I've watched people spend

a week writing a plan, right? I've

watched them spend two weeks. I've

watched people spend eight weeks writing

a plan at times. The plan ends up being

a hedge and it ends up being much more

expensive than the product will ever be.

And ironically, the plan is almost

always incorrect. It almost always never

survives contact with reality. It's a

truism in business. I'm just going to

tell you, look at the planning that

you're doing and see if you can cut the

planning down. See if you can cut it. I

don't know. Set a goal 90%. Can you cut

your planning down by 90%. You probably

can. See if you can replace that with

learning through prototyping. See if you

can replace that with a bold rough

direction and aggressively shipping and

optimizing for what works. If you

haven't built something in the last

couple of weeks at work or maybe at home

if you're doing a home project, then

you're probably overplanning. Let

reality inform the plan instead of

trying to predict reality with the plan.

Because prediction, ironically,

prediction has now become expensive and

luxurious because execution and doing is

cheap and execution and doing is more

accurate and more reliable. Learning

from customers directly by launching is

more reliable. And so think about how

much you're planning and see if you can

cut it by 90%. I'm not kidding. See if

you can cut it down and see if you can

spend that energy on doing instead.

Number six, the deck instead of the

demo. That's an old habit, right? We

build consensus and we create a deck and

we make sure stakeholders can weigh in

and we we have we call them a walking

around deck. Does that ring a bell for

you? The deck shows what we want to

envision. It helps us to build uh

alignment with various stakeholders.

Sometimes we workshop that deck. Some I

remember having to pick the fonts on

that. Forget all of that. Build the

demo. Show it. Does it does this sound

like I'm beating the same drum? It

should build a working prototype and

show that instead of the deck. In some

senses, work is getting much simpler

now. All of the rituals that we prepared

as a hedge against execution are in

question. We should be asking if we need

them. Why not just do the work? Why not

just ship to customers? Habit number

seven, consensus before action. Get

everybody aligned. This is so deadly.

And I know organizations tried to push

on this by giving individual team

leaders autonomy. Amazon had their two

pizza teams. There were other versions

of that. The difference now is that the

cost of consensus has 10xed or 100xed.

If consensus was expensive before in the

2010s, it is priceless now. And

ironically, consensus often wasn't real

anyway. People would often agree in

meetings then undermine the decisions

later. But regardless, it is too

expensive now to practice the habit of

consent. Let consensus come from the

results that create alignment. Just try

things. And again, this is going to be a

leadership change as well as an

individual contributor change. Nobody is

exempt here. Leaders can't just say do

this and then leave it up to people. But

at the same time, if you're starting out

on your own and you listen to this video

and you're trying to set a new way of

working, you're going to need some

support from your manager and you're

going to need to be explicit about that.

I mean, you can share this video if you

want, but lead by doing. Lead by showing

that if you set an ambitious goal, if

you set a larger vision for the

business, you don't need to get

consensus. You can just act in that

direction and let results create

alignment over time. I tried X and

here's what happened is much more

persuasive than let's agree to try X.

Run the experiment first and then

everyone will align when you get the

data because people can see what

actually works. And the thing is it's

never been faster. It's never been

faster to try that out. The last habit I

want to call out is the habit of

hoarding until ready. People who like to

not show work until it's complete. Who

think half finished work wastes other

people's time. I was told that when I

was a junior coming up. I was told,

"Don't submit this half-finish piece of

work. This isn't good enough to show

your manager yet. You need to finish it

all the way. Come on, Nate. What are you

doing?" Well, that is reversed. Now,

you're probably sitting on ideas and

drafts and prototypes until they feel

ready, which means you're getting

feedback late after you've invested in a

direction that might be incorrect.

Again,

this is a moment when we have to realize

that the costs of things are changing.

The cost of getting to a rapid version

is so cheap now. We just have to be

willing to show it. And that takes a

little bit of ego death on our part.

There's an ego component here. We have

to be willing to show some work that's a

little bit raw and unfinished. What if

people think it's bad? What if people

think it doesn't work? Those are the

reactions that you need. What if you

ship that prototype to the customer and

they don't like it? Well, this is why we

ship a lot and now we have that

information. Finding out that you're

wrong a week from now is better than

finding out that you're wrong a month

from now. And frankly, this puts a

primacy on thinking clearly because one

thing that doesn't go away like half

finished work that is all AI slop and

that you didn't put any thought to is

going to show. In fact, it creates

downstream work for your colleagues. But

halffinish work where you put some

thought in, you have a direction you

want to go in and you just want some

feedback on it and you want to show a

quick prototype. That's different. Now

you're putting thought into what's good

or what's not. The through line here is

pretty simple. All these eight habits

are riskmanagement rituals that made

sense when doing was expensive. And

those unit economics have all flipped.

And now the risk isn't wasting anybody's

execution time, engineering or

otherwise. It's wasting time on anything

that isn't doing anything that isn't

building. The permission loop is costing

you more than the thing you're asking

permission for. You get the idea, right?

The polish costs more than shipping. Let

me give you a couple examples of how

this looks in practice. The old way

might be you have an idea for improving

a process internally. You write up a

proposal. You schedule a meeting. The

meeting surfaces questions. You update

the proposal. Maybe weeks later, you get

the approval to try a pilot. The new

way, you get an idea. You spend an

afternoon building a rough version. You

show three people, two of them have

concerns that killed the idea. Good, you

found out in a day rather than a month.

Or they like it and you go ahead and

iterate and pilot and launch from there.

The whole thing took x less time. Or the

old way, you're working on a

presentation for leadership. You spend a

week on the deck. You refine the

transitions. You words smmith the

messages. You anticipate the objection.

Leadership asks questions. Now you spend

maybe two hours, maybe 20 minutes on a

rough deck. You just want to get the

alignment. You send it to your boss.

This is rough. Does the direction make

sense? She flags two problems you didn't

consider. You fixed them in 10 minutes

and the final deck took maybe an hour

and a half and it's actually better

because you got the feedback early. Stop

treating process as a prerequisite and

start treating iteration and trying to

get your ideas into contact with reality

as the process. The rough version is

going to be the gold standard because we

can get to it so much more quickly. Now,

there's a counterargument here I want to

call out. In some places, you're going

to hear quality matters in my domain. My

boss expects a plan. I'm going to get in

trouble if I just do things without

asking. That's fair. The habits I'm

describing have different impacts in

fields with high legal and compliance

risks. Let's say you're in medicine and

you have very high compliance risks. But

here's what I would ask. How much of the

process you're following is actually

required? And how much is it just the

way things have always been done in that

environment? And most of us when we're

honest realize that we probably have

more latitude than we're using. And the

habits that feel mandatory are often

just the defaults that nobody has

questioned before. And so the risk feels

significant. But if we have our eye on

quality in those fields in medicine, in

law, in finance, then we are going to be

in line with compliance even if we get

there more quickly. And if you want a

place to start, pick the habit that I

suggested, one of the eight. Pick the

one that feels lowest stakes to you,

right? And just just break that habit.

Just give it a try. Try a simpler way to

do it. Maybe it's shipping something

without the usual polish. Maybe it's

skipping a meeting and instead of a

brainstorm, you just build the thing.

Maybe it's stopping waiting. You're

blocked on a decision. Instead of

waiting, you make the decision. You say,

"This is what I'm deciding." And you do

the work. My my goal is not to get you

to break rules here. My goal is to get

you to discover that the principle of AI

native work in AI is recognizing where

value is really coming from in your

work. It's not coming from protecting

execution anymore. It's coming from

doing the execution quickly and doing it

within a framework where you have the

ambition, the boldness to build

something really meaningful, to solve

meaningful customer problems, to do what

you need to do to deliver value. The

people who figure this out first are

going to be operating at a velocity that

feels a lot more like Anthropic, a lot

more like cursor and a lot less like a

traditional big company, right? You name

your big company. And it won't be

because they have better tools because

almost everybody is getting the fancy

tools now. It's because they will have

stopped doing the things that are no

longer worth doing in a world where

execution is cheap. And so they'll be

shipping while other people plan.

They'll be iterating while other people

align. They'll be learning while other

people polish. If we circle back to the

beginning of this video, I talked about

this idea that we live in a world of

chaos. And we need a simplifier. The

chaos you're feeling is not random. It's

the gap between where the bottleneck has

moved and the habits you still have

today. When you close that gap, when you

start to align your work habits to how

AI actually changes scarcity in the

business to how AI enables execution and

you recognize where where there's other

scarcity in the business that needs your

business judgment, suddenly you're going

to know where to spend your time. You're

going to know why you need to move

faster and the chaos is going to start

making sense. You're going to understand

why meetings can't be the default

anymore. You're going to understand that

the next time someone ships a vibecoded

this or that, it's not just one more

piece of AI news. It's someone who

recognized that they needed to get their

idea into contact with reality quickly.

And you're going to recognize the truly

precious resources that are good for

your career and that help businesses

thrive are still going to be true in the

age of AI and are even more true now

when execution is cheap. Things like

clarity, like ambition, like

distribution. the bottlenecks that are

appearing and growing more and more

scarce as execution driven by AI gets

more and more pervasive. And so in a

world where we're going to get, I

guarantee it, another major AI release

tomorrow, worry less about what

execution is enabling a company to do,

and worry more about your ability to

shift your work habits and your ability

to practice getting your ideas into

contact with reality, solving real

customer problems, and shifting your

attention to the things that are

actually difficult to do, the things

that require good human judgment. It's

hard to get clear on things. It's hard

to be ambitious. We tend to think

smaller than we should. Distribution is

a hard thing to tackle. Most

entrepreneurs actually overindex on

product and underindex on go to market.

That's not new. But distribution is

exponentially more valuable now. So

that's my challenge to you. In a world

that is chaotic, recognize that so much

of the chaos comes from being out of

sync with where AI is pushing scarcity

in your business. It feels out of sync

because AI is making execution so cheap

and everything is changing because of

that. Let's get that figured out and

things are going to start to get

smoother.