AI Made Me

Better. Faster.

An honest look into the strange new life

of a developer in the AI era.

Exhausted.

Exhausted.

Anca Spătariu

  • Core Software Engineer @ WeVideo

  • Community Lead @ tim.js

  • NOT an AI expert

This is NOT an anti-AI talk

This is NOT an anti-AI talk

I use AI. I like AI.

I also have questions...

THE PROMISE

Part I

AI is truly amazing

AI is good at fixing blank page syndrome

AI is good at fixing blank page syndrome

  • Don't know where to start?
  • Need a quick MVP?
  • Don't want to waste time with boilerplate?
  • Explore a new API
  • Write an RFC

AI is good at removing unpleasant work

AI is good at removing unpleasant work

-15%

Lower burnout scores when AI reduces repetitive work

Source: Boston Consulting Group Survey, January 2026 / HBR

AI feels like a superpower

AI feels like a superpower

debug

FASTER

AI feels like a superpower

debug

learn

FASTER

AI feels like a superpower

debug

learn

prototype

FASTER

FASTER

FASTER

But it comes with side effects

But it comes with side effects

  • workload creep
  • blurred boundaries
  • decision fatigue

THE TRAP

Part II

We voluntarily did more

Why did you get into programming?

Why did you get into programming?

THE

CRAFT

THE END

PRODUCT

ME

"I don't use AI, I use my brain."

Someone told me

Using AI is not less thinking.

Using AI is not less thinking.

It is different thinking.

Before:

Write code → Debug → Review → Own

Now:

Frame → Prompt → Verify → Decide → Own

Allegedly a page from an internal IBM training in 1979. Original source not confirmed

Before:

Write code → Debug → Review → Own

Now:

Frame → Prompt → Verify → Decide → Own

AI Generates. Humans own.

A day in life

2020 / 2026

A day in life

2020

Planning / Complex tasks

Execution / Simple Tasks

20%

80%

Perceived, not measured

A day in life

2026

Planning / Complex tasks

Execution / Simple Tasks

80%

20%

Perceived, not measured

Let's imagine the automation in a call center

An analogy used by psychologist Dr. Tara Behrend, 
at a Feature Stage session at APA 2025 
An analogy used by psychologist Dr. Tara Behrend, 
at a Feature Stage session at APA 2025 

Let's imagine the automation in a call center

Routine call 😊

Angry customer 😡

A day before AI automation

💁‍♀️

An analogy used by psychologist Dr. Tara Behrend, 
at a Feature Stage session at APA 2025 

Let's imagine the automation in a call center

Routine call 😊

Angry customer 😡

💁‍♀️

A day after AI automation

An analogy used by psychologist Dr. Tara Behrend, 
at a Feature Stage session at APA 2025 

Let's imagine the automation in a call center

Routine call 😊

Angry customer 😡

💁‍♀️

A day after AI automation

🤖

Increased your productivity by 20%?

Increased your productivity by 20%?

Congrats, you have bandwidth now!

AI made it feel possible

AI made it feel possible

So we volunteered

Enter: Task expansion

Enter: Task expansion

Or when "Not my job" becomes... doable? 🤔

AI did not replace me, it promoted me without asking.

More code ≠ more progress

More code ≠ more progress

Generated code:

Review capacity:

+ review tools

Human

The bottleneck moved

The bottleneck moved

Generate

Review

Test

Integrate

Maintain

THE COST

Part III

Brain fry, boundaries, and mental load

How many cores does your brain have?

How many cores does your brain have?

Code

Slack

Code review

Refinement

Code

Code

How many cores does your brain have?

AI output #1

AI output #2

Slack

Code review

Code review

Code review

Code review

Architecture

All hands

Refinement

AI output #3

Welcome to BRAIN FRY

Welcome to BRAIN FRY

 The mental fatigue from excessive AI use or oversight
beyond your cognitive capacity

Symptoms of BRAIN FRY

Symptoms of BRAIN FRY

  • brain fog
  • slower decisions
  • buzzing
  • headaches
  • overwhelm

Oversight is work

Oversight is work

+ 14%

+ 12%

+ 19%

mental effort

mental fatigue

information overload

Source: HBR, “When Using AI Leads to Brain Fry”, 2026

How many AI tools is too many?

How many AI tools is too many?

How many AI tools is too many?

1      

2

3

4+

Number of tools used at once

Productivity rating 1-5

Source: Boston Consulting Survey, January 2026 / HBR

3

3.5

4

4.5

Can't stand idle agents?

Can't stand idle agents?

  • Just one prompt during lunch
  • Just one check during coffee
  • Sneaking prompts in meetings, while other people talk

Next thing you're feeding your agent after midnight

But your brain needs recovery

Work becomes ambient

Work becomes ambient

Always available.

Always possible.

Always running.

Prompts feel harmless

The productivity high

The productivity high

Fast feedback feels good

Fast feedback feels good

But sometimes... It doesn't help you work

But sometimes... It doesn't help you work

It helps you find more work

Protect your reliable / "go-to" developers at all cost

Protect your reliable / "go-to" developers at all cost

Yes, especially if that's you

  • more scope
  • more reviews
  • more meta work
  • less recovery

Protect your reliable / "go-to" developers at all cost

You may think all you need are some

But the system was rigged way before AI

But the system was rigged way before AI

The most capable people are

the first to burn out

Mental health is not a personal optimization problem

Mental health is not a personal optimization problem

You can't pomodoro 🍅 your way out of this

Companies need norms and standards around AI use

Companies need norms and standards around AI use

A human-centered AI working protocol

THE CRAFT

Part IV

Expertise matters more than ever

AI makes "I can start this" sound like "I know what I'm doing"

AI makes "I can start this" sound like "I know what I'm doing"

The illusion of competence

The less domain knowledge you have,

the harder it is to judge AI output

The less domain knowledge you have,

the harder it is to judge AI output

LGTM 🚀

Ignored patterns

Missing business context

Causes regressions

Expertise becomes the filter

Expertise becomes the filter

🤖

👩‍💻

Fluent output

Fluent output

Weak judgement

Strong judgement

+

+

=

=

Liability 😖

Leverage ✅

The manual QA revenge arc

Look who's relevant now

People ask AI

Not mentors, not the community

Sources: Stack Overflow Developer Survey 2025; Pragmatic Engineer

The questions did not disappear, they moved

The questions did not disapear, they moved

Public learning

Private learning

  • Meetups like tim.js
  • Blogs
  • Twitter
  • Stack Overflow
  • Lunch and learns
  • ChatGPT
  • Claude
  • Copilot
  • Perpexity
  • etc.

Don't get me wrong: AI for learning is AMAZING

Don't get me wrong: AI for learning is AMAZING

But we need to protect the human spaces where shared standards and communities are built

Community is still infrastructure

Not for answers. For context.

THE FIX

Part V

How do we use AI without becoming exhausted?

So what do we do?

Photo source: technologyreview.com

Not less AI.

More intentional use of AI.

What every individual can do

 Protect your attention 

Your brain is the bottleneck

 Protect your attention 

Personal AI Protocol:

  • Just one deep thread
  • Learn to notice the loop
  • Explain it to yourself

Your brain is the bottleneck

 Protect your attention 

Use AI as an adversary

Use AI as an adversary

  •  What am I missing?
  • Where can this fail?
  • Argue against this approach.

Resources

Adversarial review - Codex plugin for Claude Code

Grill me - a skill to stress-test the plan before building

🔗 https://www.aihero.dev/my-grill-me-skill-has-gone-viral

🔗 https://community.openai.com/t/introducing-codex-plugin-for-claude-code/1378186

Protect the time AI saved

Protect the time AI saved

saved time ≠ free capacity

What staff/principal engineers can do

Lead by example

Lead by example

Don't just tell people to use AI.

Coach them on how to use it safely.

Make review load visible

Make review load visible

Code reviews are often invisible work.

But real work nonetheless.

What companies/managers 

can do

Don't measure activity

Don't measure activity

I mean it. Don't.

Outcome vs. Output

Outcome vs. Output

Tokens are not impact.
Lines of code are not value.

Clarify the deal

Clarify the deal

When AI saves time, where does that time go?

AI adoption must include workload expectations!

Managers need training too

Managers need training too

Mental fatigue

Supported

teams

Figure it out

yourself

Managers need training too

Supported

teams

Figure it out

yourself

Mental fatigue

-15%

Managers who answer AI questions →

15% lower mental fatigue

Source: HBR, “When Using AI Leads to Brain Fry”, 2026

Not just governance.

A human-centered AI working protocol.

PAUSE

PACE

GROUND

Not just governance.

A human-centered AI working protocol.

PAUSE

PACE

GROUND

Not just governance.

A human-centered AI working protocol.

before important decisions

PAUSE

PACE

GROUND

before important decisions

protect focus, avoid constant reacting

Not just governance.

A human-centered AI working protocol.

PAUSE

PACE

GROUND

before important decisions

protect focus, avoid constant reacting

in human conversation

Not just governance.

A human-centered AI working protocol.

Humans are not the bottleneck.

Humans hold the judgement.

Use AI like a power tool. Not a nervous system.

Thibault Sottiaux, Tech Staff @ OpenAI

THANK YOU

Have anything to ask or add?

Your feedback means more than you think!

AI made me better. Faster. Exhausted.

By Anca Spatariu

AI made me better. Faster. Exhausted.

  • 11