DEV Community: Hainan Zhao

AI Can Write Code. But Can It Feel What Your User Feels?

Hainan Zhao — Thu, 16 Apr 2026 00:51:29 +0000

Here's a line I keep coming back to:

Anything that can be spec'd out, AI will replace.

Write a login flow? Spec'd out. AI handles it.
Build a REST API with pagination? Spec'd out. Done in seconds.
Implement a sorting algorithm? Spec'd out decades ago.

The list of things that can be cleanly specified is enormous, and AI is eating through it fast. If you can write a detailed requirements document, AI can probably build it.

But here's the gap: most software that fails doesn't fail because the code is wrong. It fails because the experience is wrong.

The Photo Studio Lesson, Again

Remember the photo studios from the last post? The ones that survived the digital camera revolution?

Here's something I didn't mention. When cameras got cheap, a lot of people thought they could take their own family portraits. They bought DSLRs, set up backdrops, tried to replicate what the studio did.

The results were technically fine. Proper exposure. In focus. Everyone facing the camera.

But they felt wrong. The lighting was flat. The poses were stiff. Nobody looked relaxed. The photo was technically correct but emotionally dead.

Because the studio photographer wasn't operating a camera. They were reading the room. Noticing when someone's smile looked forced. Adjusting the light to soften a face. Telling a joke to make a child laugh naturally. Positioning bodies so a family looked like they loved each other.

The photographer was designing an experience. The camera was just the tool.

Software Is Experienced Through Human Senses

When you use an app, you're not reading code. You're seeing it, hearing it, feeling it through the interface.

You see the layout and instantly judge whether it's trustworthy.
You hear a notification sound and feel either pleased or annoyed.
You tap a button and the response time — even 200ms vs 50ms — creates a gut feeling of quality or cheapness.

These are human sensory reactions. They happen before conscious thought. You can't spec them out because they're not logical — they're embodied.

AI has never been confused. It has never stared at a screen unable to find the "save" button. It has never felt a surge of frustration when a form resets itself. It has never felt delight when an animation is perfectly smooth.

AI can optimize for metrics. But it can't feel what your user feels.

Why UX Is the Last Moat

Think about the apps you love vs. the apps you tolerate.

Instagram in the early days — the magic wasn't the photo filters. It was the feel. Open the app, snap, pick a filter, share. Three taps. The entire flow was designed around a feeling: "I just captured something beautiful and I want to share it NOW."

Superhuman — an email client that charges $30/month in a world of free Gmail. Why? Because it feels fast. Every animation, every keyboard shortcut, every transition is designed to make you feel like you're moving at the speed of thought.

TikTok — the swipe gesture, the instant full-screen video, the algorithmic feed. The entire UX is designed around one feeling: "one more." It bypasses your rational brain and goes straight to dopamine.

None of this is in a spec. You can't write "make it feel delightful" in a requirements document. It requires someone who can imagine themselves in the user's body and feel what they feel.

The Spec-able vs. The Sens-able

Here's the dividing line:

Spec-able (AI does this)	Sens-able (Humans must do this)
Build a responsive layout	Choose a layout that feels calm vs energetic
Implement form validation	Decide the error message tone — helpful vs punitive
Create a color palette	Feel whether a color combination is trustworthy
Optimize page load time	Know that 200ms feels instant but 500ms feels slow
Build a notification system	Understand that buzz at 11pm feels invasive
Implement a checkout flow	Sense where users feel anxious about spending money
Generate an icon set	Know which icon feels like "delete" vs "archive"

The left column is engineering. The right column is design. AI is demolishing the left column. The right column is untouchable — because it requires lived human experience.

What This Means for Software Teams

The shift is already happening. The most valuable person on a software team is no longer the one who writes the most code. It's the one who can:

See through the user's eyes. Look at a screen and feel "this is confusing" before the user even reports it. Not because of analytics, but because of empathy.

Hear the silence. Notice when a user journey has a dead end — a moment where the user thinks "now what?" — that no error handler will catch.

Feel the friction. Sense that a three-step flow could be one step. That a confirmation dialog feels like distrust. That loading spinners on every interaction make the app feel broken even when it's working.

Design for emotion. Not just usability, but how the product makes people feel. Confident? Relaxed? Playful? In control? That feeling IS the product.

The Studio Photographer's Instinct

Go back to the photo studio. The best photographers have an instinct — a sense — for what makes a great portrait. They can't always explain it. It's not a rulebook. It's years of observing humans, understanding light, reading emotion.

The best UX designers have the same instinct. They can look at a wireframe and feel "this won't work." They can use an app for 30 seconds and tell you exactly where users will abandon. Not because of data, but because they've trained themselves to feel what users feel.

AI can generate a thousand wireframes. It can A/B test a million variants. But it can't sit in a room with a confused user and feel the embarrassment of not understanding a simple task.

It can't feel.

And in a world where code is free and features are commoditized, feeling is the only thing that differentiates great software from adequate software.

The Future: AI Builds, Humans Feel

The software team of the future isn't bigger. It's different.

Instead of 10 engineers writing code, you'll have 2 engineers directing AI with 1 designer who owns the experience. The engineers handle the spec-able parts. The designer handles the sens-able parts.

The code will be fine. AI ensures that.

But does the app feel right? Does it make people feel smart, capable, in control? Does it respect their time, their attention, their emotions?

That's not a spec. That's a senses check.

And only a human can do it.

What app makes you feel something — good or bad? I'd love to hear about the software experiences that hit you on a sensory level.

How AI Levels the Playing Field for Small Teams

Hainan Zhao — Tue, 14 Apr 2026 11:09:25 +0000

A friend of mine works at a small milk tea chain. Twenty stores, growing fast, loyal customers. They have exactly two software engineers building their app.

The app sucks.

Orders get lost. The loyalty program breaks. The menu loads slowly. Push notifications go out at 3 AM. Customers complain in the store, and my friend just apologizes — he knows the engineers are trying their best, but two people can only do so much.

It's not that they don't want to build great software. They literally can't afford more engineers. And the off-the-shelf solutions don't fit — a milk tea chain has weird specific needs. Custom drink modifiers (50% sugar, less ice, extra boba, add cheese foam). Multi-store inventory for perishable ingredients. A loyalty program that actually works for walk-in teenage customers who don't want to sign up for anything.

This isn't a rare story. It's everywhere. Small businesses stuck with bad software, not because the problems are hard, but because they could never afford to solve them properly.

AI is changing that. And it's not just helping small businesses build better internal tools — it's creating an entirely new category of software companies that serve markets nobody cared about before.

The Cost Wall That Held Everyone Back

For two decades, software had a minimum viable budget. Below a certain threshold, it didn't matter how good your idea was — the economics didn't work.

Want to build a SaaS product? You needed a backend developer, a frontend developer, a designer, a DevOps person, a QA tester, maybe a PM. At least 12-18 months. $500K–$2M in funding.

Want to build an app for your business? Same problem. You'd hire an agency for $200K, get a mediocre result, and then pay maintenance forever. Or you'd hire two engineers and watch them drown.

This created a strange world where:

Millions of small businesses used Excel for everything
Entire professions had no dedicated software
People cobbled together generic tools for specialized workflows
The long tail of human problems had no software solutions

The software industry served the head of the distribution — big markets, big budgets. Everyone else made do.

What AI Actually Changed

One person with AI can now build what used to take a team of 10 six months. Not a prototype — a real, shippable product with auth, billing, database, deployment, the works.

This doesn't just make software cheaper. It changes who gets to have good software.

	Old Model	AI-Era Model
Team to build a product	10-50+	1-3
Build time	6-18 months	1-4 weeks
Funding needed	$500K-$5M	$0-$10K
SMB internal app budget	$200K+ (agency)	Near-zero (in-house)
Market size worth serving	Massive	Tiny niche

For SMBs like my friend's milk tea chain, this means their two engineers can suddenly do the work of ten. Build a proper ordering system in a sprint. Fix the loyalty program in a weekend. Add real-time inventory across all twenty stores without hiring anyone new.

For entrepreneurs, it means a product serving 500 people paying $50/month = $300K/year is now a viable business. For a solo founder with near-zero costs, that's life-changing money. For a VC-backed startup, it was never worth a meeting.

Two sides of the same coin: small teams can now build software that used to require big teams with big budgets.

The SMBs Get Better Software

Go back to the milk tea chain. With AI, those two engineers could build:

A proper ordering system that handles custom modifiers without breaking. A loyalty program teenagers actually want to use. Real-time inventory across all twenty stores so they never run out of boba on a Saturday. Push notifications timed to after-school hours instead of 3 AM.

And this pattern repeats everywhere:

The independent bookstore that can't afford a custom POS system. One tech-savvy employee with AI can build exactly what they need.

The local gym chain paying $2,000/month for generic gym software that doesn't fit their class-based model. Their one developer can build something better in a month.

The regional logistics company running everything on spreadsheets and WhatsApp. A small IT team can build a proper tracking and dispatch system.

These businesses don't need venture-scale software. They need software that fits their workflow, built by someone who understands their problems. That used to cost too much. Now it doesn't.

The Solo SaaS Companies Serve the Niches

The other side of the coin: some of those small team builders will turn their solutions into products.

Someone builds a great ordering system for their milk tea chain. Then realizes every bubble tea shop has the same problem. So they package it up, charge $40/month, and suddenly there's a SaaS product for bubble tea chains.

Not restaurants. Not cafes. Bubble tea. The modifier system, the seasonal drink rotation, the ingredient-level inventory for perishable toppings. There are maybe 50,000 bubble tea shops globally. A $40/month product = $24M TAM. Too small for Toast or Square. Perfect for a solo founder who lives in that world.

These niches are everywhere:

A scheduling tool for pet groomers. Not generic scheduling. Pet groomers. Different workflow, different terminology. 50,000 businesses globally. Perfect for one person.

Inventory management for coffee roasters. Batch tracking, bean origins, roast dates, wholesale vs. retail pricing. 10,000 potential customers. Beautiful for a solo founder.

Invoicing for freelance translators. Per-word pricing, language pairs, rush fees, agency vs. direct terms. 500,000 freelancers globally. A $15/month product = $90M TAM. Too small for Salesforce. Just right for a two-person team.

Practice management for speech therapists. Insurance coding, IEP goal tracking, parent portals. 100,000 potential users.

Before AI, none of these made economic sense. Now they're all viable.

Why Big Software Can't Serve These Markets

Why don't Salesforce or HubSpot just add these features? They can't. Not because of technology, but because of their economics.

A feature for 50,000 pet groomers generates maybe $1M/year. For Salesforce, that's rounding error. The PM who pitches it gets laughed out of the room. The engineering sprint costs more than the feature earns.

Big software companies are constitutionally unable to serve small verticals. Their cost structure demands big markets. They optimize for the average of a huge customer base, which means nobody gets exactly what they need.

A solo founder has the opposite economics. 50,000 customers at $30/month is life-changing money. Every single customer matters. The product fits the workflow exactly, because the founder lives in that world.

This Is Already Happening

Solo SaaS founders are building real businesses right now:

Pieter Levels built Nomad List, Remote OK, and PhotoAI — multiple products, one person, millions in revenue
Danny Postma built HeadshotPro — AI headshots, solo founder, profitable within weeks
Marc Lou ships a new product almost monthly — each one niche, each one profitable

The pattern: find a specific pain point, build a focused solution, charge for it. No board meetings. No fundraising. Just build, ship, earn.

What Happens to Big Software

The big companies aren't going away. Neither did department stores when boutiques opened. But the dynamics shift:

Market fragmentation. Instead of 3 CRM options, you'll have 50 — one for real estate, one for fitness studios, one for legal practices.

Price pressure. When a solo founder charges $15/month because costs are near-zero, the $299/month generalist needs to justify the premium.

Speed. A solo founder ships a feature in a day. A big company needs sprint planning, design review, QA cycles. The niche player moves 10x faster.

The "good enough" threshold. Most customers need 5 features that work perfectly for their workflow — not 200 features that half-work for everyone else.

The Future: Good Software for Everyone

Here's what excites me most.

My friend's milk tea chain gets a great app. Built by two engineers who suddenly have superpowers.

The pet groomer down the street gets scheduling software that actually fits. Built by someone who used to be a pet groomer.

The freelance translator gets an invoicing tool that understands per-word pricing. Built by a translator who got tired of spreadsheets.

For decades, good software was a luxury only big companies could afford. Everyone else made do with generic tools and workarounds.

That era is over. Small teams — whether building for themselves or building a product — can now create software that used to require a small army.

The long tail is open for business.

What software do you wish existed for your industry? I'd love to hear about the niches nobody's building for yet.

Why Photo Studios Survived the Digital Camera — And What It Means for Software Engineers

Hainan Zhao — Tue, 14 Apr 2026 11:03:17 +0000

When I was young, cameras were expensive. My family went to a photo studio once a year — dress up, sit still, get our portrait taken. It was a ritual. The studio owned the expensive equipment, the darkroom, the expertise. We paid for all of it.

Then digital cameras happened. Then smartphones. Suddenly everyone had a camera in their pocket.

You'd think photo studios would disappear.

They didn't. In fact, many of them charge more now than they did 20 years ago.

Why Photo Studios Survived

Nobody paid the studio because they owned a camera. They paid for the outcome: professional lighting, flattering poses, composition, retouching. The camera was just the tool.

When everyone got a camera, the average photo became worthless. But a great photo — studio lighting, professional styling, printed and framed — became more valuable. Because it stood out against a sea of mediocre smartphone shots.

When photography was hard, any decent photo was impressive. When photography became easy, only exceptional photos mattered. Studios stopped competing on "we can take a picture" and started competing on "we'll create a moment you'll frame on your wall."

Modern studios sell the ritual. The consultation, the outfit changes, the directed session, the private reveal, the printed album, the framed canvas. The image file is almost an afterthought. People spend more at studios now because they're not buying a photo — they're buying an experience.

The Surprising Data

Here's a fact that surprised me: there are more professional photographers in the US today than a decade ago. Bureau of Labor Statistics data shows photographer employment grew from 99,000 in 2014 to 117,000 in 2023 — a 17% increase. Globally, there are over 2 million professional photographers, and the field is projected to keep growing.

Camera sales dropped 84% since 2010. But professional photography didn't shrink — it evolved. More people became professionals, but the nature of the work changed: more freelancers, fewer staff positions; more specialists (weddings, newborns, commercial, drone), fewer generalists.

The camera got cheap. The profession didn't die. It upgraded.

The Software Engineer Parallel

AI is the digital camera. Code generation is the photo.

Just like digital cameras didn't kill photography — they transformed it — AI won't kill software engineering. It will transform it.

What Gets Commoditized

"Write me a CRUD API" — the equivalent of taking a photo of your lunch. Instant, free, worthless.
Boilerplate, simple scripts, standard integrations
Junior-level implementation tasks
Code that follows a pattern anyone can look up

If your value is "I can write code," you're the person with a camera phone in a world where everyone has one.

What Gets More Valuable

System design. AI can write code, but can it decide what to build? Architecture decisions, tradeoffs, scalability, security boundaries — this is the lighting and composition of software. The camera doesn't choose where to point it.

Problem framing. Clients don't come with clean specs. They come with messy, contradictory, emotional problems. Translating ambiguity into a system someone can build — that's the posing and directing. It's the hardest part, and AI is terrible at it.

Taste and judgment. When code is free, which code matters. Choosing the right abstraction. Saying no to the wrong feature. Knowing when not to build something. Recognizing that a "simple" request has implications across five systems.

Ownership of outcomes. Stakeholders don't want code. They want someone who understands their domain, communicates clearly, ships reliably, and takes responsibility when things break. This is the studio experience — the trust, the process, the delivery.

The camera got cheap. The profession didn't die — it upgraded. Photographers went from operating equipment to directing shoots, styling sets, creating experiences.

Software engineers face the same upgrade. The question isn't whether you'll have a job. The question is whether you'll be the one who tells AI what to build — or the one who gets replaced by it.

The studios didn't all die — the bad ones did. Studios that competed on "we own a camera" disappeared. Studios that sold portraits, memories, experiences raised their prices.

Same pattern: engineers who compete on "I can write code" will struggle. Engineers who sell "I understand your problem and will build the right thing" will charge more than ever.

The camera operators are already being replaced. The portrait artists are just getting started.

What do you think? Are you the camera operator or the portrait artist? I'd love to hear your perspective.

[Updated] Why Photo Studios Survived the Digital Camera — And What It Means for Software Engineers

Hainan Zhao — Tue, 14 Apr 2026 10:52:12 +0000

This post has been updated with corrected data and merged with a follow-up piece. Read the full version here.

AI Crisis Survival Guide: How Countries and Citizens Can Navigate 2026-2036

Hainan Zhao — Wed, 18 Mar 2026 15:39:33 +0000

The technology isn't the problem. The distribution of gains is.

The Reality

My previous post laid out the crisis scenarios:

India: 70-75% conflict probability (revolution/state failure)
China: 40-50% internal + 40-50% external conflict
USA: 50-60% serious unrest
EU: 20-30% (Southern Europe highest)
Japan: 10-15% (lowest risk)

This post answers: What can we DO about it?

Both as countries and as individual citizens.

Part 1: Country-Level Solutions

🇮🇳 India: Preventing State Failure

The Crisis: 200M+ unemployed youth, dual migration failure, climate stress, weak state capacity.

What Government Must Do:

Priority	Action	Timeline	Feasibility
1. Rural job guarantee	Expand MGNREGA (rural employment) to 200 days/year, include climate adaptation work	Immediate	HIGH (existing program)
2. Manufacturing push	Special economic zones, labor reform, China+1 supply chain capture	2-5 years	MEDIUM
3. UBI pilot → scale	Start with poorest districts, fund via AI/tech company taxes	1-3 years	MEDIUM
4. Climate adaptation	Drought-resistant crops, water infrastructure, rural insurance	2-5 years	HIGH
5. Education reform	Shift from IT/services to manufacturing + AI-resilient skills	3-5 years	LOW (politically hard)

What Would Actually Work:

Leverage "China+1" moment — Western companies leaving China need alternative manufacturing hubs. India can capture this IF labor laws simplify.
Rural climate jobs — Pay rural workers for water conservation, reforestation, soil restoration. Addresses unemployment + climate simultaneously.
Tech tax for UBI — Tax AI/IT companies (who benefit most from displacement) to fund basic income.

Probability of Success: 30-40% (state capacity is the limiting factor)

🇨🇳 China: Avoiding Elite Fracture + External War

The Crisis: Reverse migration, regional unrest, Taiwan flashpoint, demographic collapse.

What Government Must Do:

Priority	Action	Timeline	Feasibility
1. Regional transfers	Redirect coastal tax revenue to manufacturing provinces	Immediate	HIGH (state capacity exists)
2. UBI-lite expansion	Expand existing social safety net to displaced workers	1-2 years	HIGH
3. Taiwan de-escalation	Quiet diplomacy, delay unification timeline	Immediate	MEDIUM
4. Elderly care AI	Deploy AI/robotics for aging population (productivity gain)	2-5 years	HIGH
5. Immigration reform	Allow skilled immigration to offset demographic decline	3-5 years	LOW (culturally resisted)

What Would Actually Work:

State capacity is China's advantage — Can mobilize resources faster than democracies. Use it for regional redistribution BEFORE unrest spreads.
Delay Taiwan — External war during internal crisis = regime collapse risk. Quietly delay unification timeline.
AI for aging — China's demographics are worse than AI displacement. AI/robotics for elderly care = productivity gain + social stability.

Probability of Success: 50-60% (authoritarian systems can act fast, but elite fracture is the wild card)

🇺🇸 USA: Preventing Class War

The Crisis: Polarized politics, weak safety net, gun violence, rural/urban split.

What Government Must Do:

Priority	Action	Timeline	Feasibility
1. Federal job guarantee	Infrastructure, care work, climate adaptation jobs	Immediate	LOW (politically polarized)
2. Wealth tax	Fund UBI/social programs via top 1% wealth tax	1-3 years	LOW (elite resistance)
3. AI dividend	Citizen ownership of AI gains (Alaska Permanent Fund model)	2-5 years	MEDIUM
4. Mental health infrastructure	Address depression/suicide crisis from displacement	Immediate	MEDIUM
5. Political reform	Ranked-choice voting, anti-gerrymandering, reduce polarization	3-5 years	LOW (entrenched interests)

What Would Actually Work:

FDR Coalition 2.0 — Only cross-class coalition (workers + middle class + some elites) can pass redistribution. Requires charismatic leadership.
State-level experiments — Blue states implement UBI/AI dividends first. Proof of concept forces federal action.
Corporate capture — Tech companies may SUPPORT UBI (need consumers with purchasing power). Align corporate + worker interests.

Probability of Success: 40-50% (political polarization is the limiting factor)

🇪🇺 EU: Preventing Fragmentation

The Crisis: North/South split, aging, migration pressure, coordination failure.

What Government Must Do:

Priority	Action	Timeline	Feasibility
1. EU-wide AI dividend	Distribute AI gains across all member states	2-5 years	MEDIUM
2. Nordic model expansion	UBI pilots → national → EU-wide	3-5 years	HIGH (Nordic leads)
3. Southern Europe investment	Infrastructure, jobs in Italy/Spain/Greece	2-5 years	MEDIUM
4. Managed migration	Legal pathways + integration programs	Immediate	LOW (politically toxic)
5. Pension reform	Address aging crisis before fiscal collapse	Immediate	LOW (elderly voters resist)

What Would Actually Work:

Nordic proof of concept — Sweden/Denmark/Finland implement UBI first. Success forces Southern adoption.
EU cohesion funds 2.0 — Redirect existing EU funds to AI displacement hotspots (not just poor regions).

Probability of Success: 50-60% (stronger institutions than US/India, but coordination is slow)

🇯🇵 Japan: Managed Decline

The Crisis: Aging, shrinking GDP, immigration resistance.

What Government Must Do:

Priority	Action	Timeline	Feasibility
1. Robotics deployment	AI/robots for elderly care, services	Immediate	HIGH
2. Skilled immigration	Targeted visas for care workers, tech	1-3 years	MEDIUM
3. Elderly workforce	Raise retirement age, flexible work	Immediate	HIGH
4. Productivity focus	AI for GDP quality (not quantity)	2-5 years	HIGH
5. Cultural adaptation	Shift from growth mindset to stability mindset	5-10 years	MEDIUM

What Would Actually Work:

Robotics leadership — Japan already leads. Double down on AI/robots for aging society.
Immigration pragmatism — Frame as "skilled workers" not "immigrants." Culturally palatable.

Probability of Success: 70-80% (most manageable crisis of all major countries)

Part 2: Individual Survival Guide

Universal Strategies (All Countries)

1. Build AI-Resilient Skills

Skill	Why It's Resilient	How to Learn
Care work (elderly, child, health)	Human touch hard to automate	Nursing, therapy, coaching certifications
Creative work (art, storytelling, design)	Human creativity valued	Portfolio building, niche specialization
Community building (organizing, events, local governance)	Human connection irreplaceable	Volunteer, lead local organizations
AI oversight (evaluation, auditing, ethics)	Humans must oversee AI	AI ethics courses, auditing certifications
Skilled trades (electrician, plumber, HVAC)	Physical + problem-solving hard to automate	Trade schools, apprenticeships

AVOID:

Data entry, basic coding, call centers, routine analysis (all AI-displaceable)
Middle management (coordination increasingly automated)
Any job that's mostly screen-based + predictable

2. Financial Preparation

Action	Why	How
Reduce debt	Unemployment = can't service debt	Pay down high-interest debt NOW
Diversify income	Single job = single point of failure	Side gigs, passive income, skills monetization
Build 12-month emergency fund	Standard 6 months isn't enough	Automate savings, cut discretionary spending
Own productive assets	Land, tools, skills > paper wealth	Community gardens, maker equipment, rental properties
Geographic arbitrage	Some regions hit harder than others	Consider moving to lower-cost, AI-resilient areas

3. Community Building

Why: When safety nets fail, community is your backup.

Action	Impact	How to Start
Join local organizations	Mutual aid, information sharing	Religious groups, hobby clubs, neighborhood associations
Build neighbor relationships	Emergency support network	Introduce yourself, share contact info, organize block parties
Skill-sharing networks	Trade services without money	Time banks, skill swaps, tool libraries
Political organizing	Demand redistribution BEFORE crisis	Join advocacy groups, attend town halls, vote

4. Mental Health Preparation

Why: Displacement = identity crisis + depression.

Action	Why	How
Separate identity from job	Job loss = less devastating	Hobbies, community roles, family identity
Build resilience practices	Handle uncertainty better	Meditation, therapy, exercise, journaling
Maintain routines	Stability during chaos	Sleep, meals, exercise, social connection
Limit doomscrolling	Anxiety without action = paralysis	Set news limits, focus on local action

Country-Specific Individual Strategies

🇮🇳 India: Survival Priorities

Highest Risk: Urban IT workers, rural landless laborers, young graduates.

Do This:

If in IT/services: Start transitioning NOW (2026-2028 window). Move to:
- AI oversight/evaluation roles
- Manufacturing management (China+1 boom)
- Care work (growing domestic demand)
- Government jobs (more stable)
If rural: Don't migrate to cities yet. Wait for:
- Rural job guarantee expansion
- Climate adaptation programs
- Local manufacturing zones
If young graduate: Consider emigration (Gulf, Europe, Canada) OR government jobs.
Financial: Hold gold, land, foreign currency (rupee volatility risk).
Community: Caste/religious networks will matter MORE during crisis. Strengthen these ties.

Don't:

Take on large debt (education loans, mortgages)
Bet everything on IT career
Move to megacities without job security

🇨🇳 China: Survival Priorities

Highest Risk: Manufacturing workers, coastal migrant workers, young urban professionals.

Do This:

If manufacturing: Return to hometown BEFORE mass layoffs (2027-2029). Rural land = safety valve.
If urban professional: Build guanxi (relationships) with local officials. Connections matter during crises.
If young: Consider emigration (Southeast Asia, Europe) OR state sector jobs (more stable).
Financial: Diversify outside China (Hong Kong accounts, foreign assets if possible).
Community: Family networks are critical. Maintain strong ties.

Don't:

Protest openly (crackdown risk)
Bet on coastal real estate (regional crisis risk)
Assume CCP will bail you out (regional variation in response)

🇺🇸 USA: Survival Priorities

Highest Risk: White-collar knowledge workers, gig workers, rural communities without industry.

Do This:

If knowledge worker: Transition to AI-resilient roles (see skills above). Don't wait for layoff.
If gig worker: Unionize + diversify. Single platform = single point of failure.
Geographic: Consider moving to:
- Blue states with stronger safety nets (CA, NY, MA)
- Rural areas with low cost of living + community ties
- AVOID: Rust Belt 2.0 (manufacturing regions without diversification)
Financial: Max out retirement accounts NOW. Diversify across asset classes.
Community: Join local organizations BEFORE crisis (churches, unions, hobby groups, mutual aid).

Don't:

Live paycheck to paycheck (build 12-month emergency fund)
Isolate politically (cross-partisan relationships matter)
Assume federal government will save you (state/local action more likely)

🇪🇺 EU: Survival Priorities

Highest Risk: Southern Europe youth, migrant communities, gig workers.

Do This:

If Southern Europe: Consider moving to Nordic/Western Europe (work opportunities).
If migrant: Strengthen legal status NOW. Crisis = scapegoating risk.
If young: Learn German/Nordic languages (better job markets).
Financial: EU-wide banking diversification (don't keep everything in one country).
Community: Cross-border networks matter (EU freedom of movement is your advantage).

Don't:

Assume EU will bail out your country (coordination is slow)
Stay in declining regions without exit plan
Ignore far-right rise (affects migrants/minorities first)

🇯🇵 Japan: Survival Priorities

Highest Risk: Young workers (shrinking opportunities), elderly without family.

Do This:

If young: Learn skills for aging society (healthcare, robotics, elderly care).
If elderly: Build community ties (isolation = high mortality risk).
Financial: Japan's debt is sustainable... until it isn't. Diversify some assets overseas.
Community: Neighborhood associations (chonaikai) are critical. Participate.

Don't:

Assume lifetime employment (it's already gone)
Isolate (community is your safety net)
Bet everything on Japanese stocks/bonds (demographic headwinds)

The Bottom Line

For Countries:

Reform is possible. But requires:

Political will (cross-class coalitions)
Timing (BEFORE crisis peaks)
Resources (tax AI winners, redistribute)

Window: 2026-2030. After 2030, crisis momentum is hard to stop.

For Individuals:

You have more agency than you think.

Timeline	Action
Now (2026)	Build emergency fund, reduce debt, start skill transition
2026-2028	Complete skill transition, strengthen community ties
2028-2030	Geographic positioning (if needed), political organizing
2030+	Ride out crisis with preparation + community support

The goal isn't to "beat" the crisis. It's to:

Survive with your family intact
Maintain mental/physical health
Come out the other side with community stronger

Your ancestors survived wars, depressions, pandemics, and collapses.

You have advantages they didn't: information, mobility, technology, global networks.

Use them wisely.

The crisis is coming. But fatalism is a choice. Preparation is also a choice.

Choose preparation. Build community. Demand better from leaders.

What's your situation? Which country, what work, what's your plan? Drop a comment — let's share strategies.

Published: March 18, 2026

Part 1: AI, Unemployment, and War: A Country-by-Country Simulation

AI, Unemployment, and War: A Country-by-Country Simulation (2026-2046)

Hainan Zhao — Wed, 18 Mar 2026 11:05:58 +0000

Mass unemployment without safety nets = revolution or war. This isn't speculation — it's historical law.

The Question Nobody Wants to Ask

What happens when AI displaces 40-50% of jobs before UBI exists?

Not in 2050. In the next 10-15 years.

History gives us one answer: violence.

When people can't feed their families, when they see elites thriving, when there's no legitimate path forward — revolution and war become rational choices.

Let me show you what this looks like by country.

The Breaking Point Math

Variable	Current Trajectory	Breaking Point
AI displacement	20-30% by 2035	25%+ unemployment
UBI implementation	Pilot programs only	Needs 30%+ coverage
Wealth concentration	Top 1% = 45% global wealth	Top 1% = 60%+
Youth unemployment	15-25% (varies)	40%+

Historical precedent: When youth unemployment hits 40%+ AND inequality peaks AND no safety net exists → conflict becomes probable.

Event	Unemployment	Safety Net	Outcome
Great Depression (1930s)	25% US	None initially	New Deal OR WWII
Weimar Germany (1920s-30s)	30%+	Collapsed	Nazi rise → WWII
Arab Spring (2010-12)	25-30% youth	Weak/Corrupt	Revolutions
Rust Belt (1980s-2000s)	15-25% regional	Partial	Political radicalization

The dangerous window: 2028-2035.

This is when AI displacement accelerates, UBI still doesn't exist, and the first fully displaced generation reaches political age.

🇺🇸 United States: Class War

The Setup

Youth population: 13% (moderate)
Political system: Democracy (polarized, gridlocked)
Safety net: Weak (no UBI, limited welfare)
Inequality: Top 1% = 35% wealth
Guns: 400M+ in circulation

Most Likely Scenario: Civil Unrest / Class Conflict (65% probability)

Timeline:

Year	Event
2028-30	AI displacement hits 20%, youth unemployment 35%+
2029-31	Mass protests in tech hubs (SF, NYC, Seattle)
2030-32	Red states vs. Blue states fracture (UBI states vs. free market)
2031-33	Armed militias form (both left and right)
2032-35	Either New Deal 2.0 OR low-level civil conflict

Key dynamics:

Gen Z/Alpha = first fully displaced generation
Rural/urban split = different displacement rates
Racial dimension = minority youth hit first/worst
Gun violence = protests turn deadly faster than other countries

Outcome:

Best case: FDR-style coalition, UBI, wealth tax
Worst case: Low-level civil conflict (sustained unrest, militia violence)
Most likely: Hybrid — unrest forces elite concession, but deep scars remain

Conflict probability: 50-60% (serious unrest), 15-20% (sustained violence)

🇨🇳 China: Authoritarian Control + External War

The Setup

Youth population: 11% (shrinking fast)
Political system: Authoritarian (CCP one-party rule)
Safety net: Moderate (urban pensions, rural weak)
State capacity: High (can mobilize resources fast)
Surveillance: Social Credit, facial recognition nationwide
Urbanization: Already 65% (most migration already happened)

Most Likely Scenario: Control + Diversionary Aggression (70% probability)

Timeline:

Year	Event
2027-29	Manufacturing AI hits 30% displacement, factory towns empty
2028-30	Reverse migration: workers return to rural hometowns (not cities)
2029-31	Rural areas can't absorb returnees → regional unrest in manufacturing provinces
2030-33	CCP response: Social Credit tightened + nationalism + external enemy (Taiwan)
2032-35	Either successful control OR elite fracture (coastal vs. interior) + unrest

Key dynamics:

NOT urban migration — urbanization already 65%, cities LIMIT migration via hukou
Reverse migration — displaced workers return to rural areas (land = safety valve)
Regional concentration — unrest in Guangdong, Zhejiang, Jiangsu (manufacturing heartland)
Local government collapse — tax revenue falls → can't pay pensions → spiral
Taiwan flashpoint — 2027-2032 window (diversionary war OR genuine unification push)

Why this is MORE dangerous than urban unrest:

Regional concentration = harder to contain than scattered urban protests
Return migration stigma = visible failure (shame + desperation)
Elite fracture risk = coastal provinces (rich) vs. interior (poor) → CCP unity strained

Outcome:

Best case: Authoritarian stability, UBI-lite, nationalism sustains regime
Worst case: Elite fracture, mass regional unrest, regime collapse
Most likely: Control holds through 2035, but external conflict risk high

Conflict probability: 40-50% (internal — revised UP from 30-40%), 40-50% (external — Taiwan/resources)

🇪🇺 European Union: Reform + Fragmentation

The Setup

Youth population: 11% (low)
Political system: Democracy (varies by country)
Safety net: Strong (welfare states, unemployment benefits)
Inequality: Moderate (Nordic low, Southern high)
Fragmentation: 27 countries, hard to coordinate

Most Likely Scenario: Two-Speed Europe (60% probability)

Timeline:

Year	Event
2028-30	AI hits services, youth unemployment 25-30%
2029-31	Nordic countries implement UBI first
2030-32	Southern Europe (Italy, Spain, Greece) = unrest
2031-33	EU-wide AI dividend debated (Germany resists, Southern demands)
2032-35	Two-speed Europe: Nordic/Western (reform) vs. Southern (unrest)

Key dynamics:

Aging crisis = bigger than AI (who pays pensions?)
Nordic model = proof of concept for UBI
Southern Europe = Greece 2010s on steroids
Migration = AI displacement + climate migration = tension

Outcome:

Best case: EU-wide AI dividend, Nordic model spreads
Worst case: EU fractures, populist takeovers, Schengen collapses
Most likely: Reform in North/West, unrest in South, EU survives but weaker

Conflict probability: 20-30% (serious unrest in Southern Europe)

🇯🇵 Japan: Managed Decline

The Setup

Youth population: 9% (tiny)
Political system: Democracy (stable, LDP dominant)
Safety net: Moderate (pensions, health, fiscally strained)
Inequality: Low (by global standards)
Aging: 29% elderly (highest in world)

Most Likely Scenario: Robotics Integration + Stability (75% probability)

Timeline:

Year	Event
2027-29	AI + robotics fill labor gaps (elderly care, services)
2028-30	Immigration debates intensify
2029-31	UBI-lite for elderly, work requirements for youth
2030-33	Society adapts: AI = solution to aging, not threat
2032-35	Stable but shrinking: GDP declines, quality of life stable

Key dynamics:

Aging = bigger issue than AI (AI solves labor shortage)
Robotics leadership = faster integration than others
Social cohesion = less unrest risk
Immigration = the real political fight

Outcome:

Best case: AI/robotics enable "soft landing" (shrinking but stable)
Worst case: Fiscal collapse, forced mass immigration
Most likely: Managed decline — GDP shrinks, society remains stable

Conflict probability: 10-15% (lowest of all major countries)

🇮🇳 India: Revolution

The Setup

Youth population: 18% (HUGE youth bulge — 200M+ people)
Political system: Democracy (fragile, populist trends)
Safety net: Weak (limited welfare, no UBI)
Inequality: Top 1% = 40%+ wealth
Tensions: Religious/ethnic (Hindu-Muslim, caste)
Urbanization: 36% (early stage — unlike China's 65%)

Most Likely Scenario: Revolution / State Failure (75% probability — revised UP)

Timeline:

Year	Event
2027-29	IT/call center jobs collapse (AI takes 50%+), 10M+ unemployed
2028-30	Dual migration crisis: Rural→urban continues (rural worse) + urban→rural reverse migration (jobs gone)
2029-31	Nowhere can absorb: Cities overflowing, rural land too fragmented to absorb returnees
2030-32	Mass protests in Bangalore, Mumbai, Delhi + rural unrest (climate + unemployment)
2031-34	Religious/ethnic scapegoating (Hindu-Muslim tensions flare)
2032-35	Revolutionary government OR authoritarian crackdown OR state fragmentation

Key dynamics:

Youth bulge = 200M+ young people with NO jobs
IT sector collapse = India's biggest success becomes biggest vulnerability
Dual migration failure:
- Rural → urban continues (climate stress, land fragmentation push migrants)
- Urban → rural reverse migration (IT workers return when jobs disappear)
- Neither direction works: Cities can't absorb, rural land too small/low productivity
Climate stress compounds: Agriculture hit by droughts/floods = rural crisis accelerates
Weak state capacity = can't implement UBI or control unrest
No manufacturing base = nowhere to retrain workers (unlike China 1990s)

Why India is MORE dangerous than China:

No rural safety valve — land already fragmented, can't absorb returnees
Weak state capacity — can't mobilize resources to crisis regions
Climate vulnerability — agriculture-dependent rural population hit hardest
Dual crisis — urban AND rural unrest simultaneously (not contained to regions)

Outcome:

Best case: Manufacturing boom absorbs youth (China 1990s model) — increasingly unlikely
Worst case: Revolution, state fragmentation, civil conflict (religious/ethnic dimensions)
Most likely: Sustained unrest, populist/authoritarian government, instability persists, mass migration to other countries

Conflict probability: 70-75% (serious unrest/revolution — revised UP), 30-40% (civil conflict/state fragmentation)

📊 Summary Table

Country	Most Likely Scenario	Conflict Probability	Timeline
🇺🇸 USA	Class War / Civil Unrest	50-60%	2028-35
🇨🇳 China	Authoritarian Control + External War	40-50% internal, 40-50% external	2027-33
🇪🇺 EU	Reform + Fragmentation	20-30%	2028-33
🇯🇵 Japan	Managed Decline + Robotics	10-15%	2027-35
🇮🇳 India	Revolution / State Failure	70-75% (revised UP)	2027-35

Highest risk: India > China > USA (China & India revised UP)

Lowest risk: Japan > EU (Nordic/Western)

Global Spillover Effects

Risk	Probability	Impact
US civil unrest → Global recession	50%	HIGH
China-Taiwan war → Supply chain collapse	35%	VERY HIGH
India state failure → 200M+ refugees + regional contagion	45% (revised UP)	CATASTROPHIC
EU fragmentation → Trade wars	25%	MEDIUM-HIGH
Multiple crises simultaneously	35% (revised UP)	CATASTROPHIC

What Prevents This?

Reforms that reduce conflict risk:

Reform	Effect	Probability by 2035
UBI	Reduces survival anxiety	40-50%
Wealth taxes	Reduces inequality	30-40%
Job guarantees	Provides purpose + income	20-30%
AI dividends (citizen ownership)	Distributes AI gains	10-20%
4-day work week	Spreads remaining work	30-40%
Universal services (housing, health)	Reduces cost of living	40-50%

Combined effect: If 3+ reforms implemented → conflict probability drops from 60%+ to 20-30%.

The Bottom Line

War/unrest without UBI: 60-70% probability.

This isn't speculation. It's historical pattern recognition.

The question isn't IF there will be unrest.

The question is:

Will it be contained (reform, UBI, redistribution)?
Or will it be catastrophic (civil war, international conflict, authoritarianism)?

The next 10 years (2026-2036) determine which path.

What Can Be Done?

For policymakers:

Implement UBI pilots → scale nationally (before 2030)
AI dividend models (citizen ownership of AI gains)
Wealth taxes (prevent extreme concentration)
Job guarantees (purpose + income during transition)

For citizens:

Political organizing (demand redistribution BEFORE breaking point)
Cross-class coalitions (workers + middle class + some elites)
Vote for candidates with concrete AI transition plans
Build community infrastructure (clubs, civic orgs, local governance)

For AI developers:

Advocate for citizen ownership models
Build transparency into AI systems
Support UBI/AI dividend research
Don't build AI that displaces without replacement pathways

The technology isn't the problem. The distribution of gains is.

Your generation decides this. Not through AI development — through political organizing, voting, coalition-building, and demanding redistribution BEFORE the breaking point.

The window is 2026-2035.

After that, history takes over.

What's your take? Too pessimistic, or about right? Drop a comment below — this conversation matters.

Published: March 18, 2026

Updated: China migration analysis corrected (reverse migration, not rural→urban). Internal conflict probability revised from 30-40% to 40-50%. India dual migration crisis added (rural + urban both failing). India conflict probability revised from 60-70% to 70-75%.

AI FOMO for Software Engineers: A 2026 Reality Check

Hainan Zhao — Tue, 17 Mar 2026 10:47:45 +0000

AI won't replace good engineers. But good engineers using AI WILL replace engineers who don't.

The FOMO Is Real

Your company announced "AI-first" strategy. Three teammates quit for "AI-native" startups. Your LinkedIn is full of new "AI Engineers."

You're wondering: "Am I falling behind?"

That's AI FOMO. And it's making engineers do stupid things.

Real FOMO Disasters I've Seen

Company	FOMO Move	Reality
Startup A	Replaced backend with "AI agents"	10x latency, 100x cost, 3 months late
Startup B	Hired 5 "AI engineers" first	Burned $2M, shipped nothing
Startup C	Built RAG for PostgreSQL data	6 months wasted, could've been SQL
Team D	Used Claude Opus for Haiku tasks	$50K/month API bill, 90% wasteful

The cost: 40-50% of AI prototypes never ship. 25-30% could've been simpler. Average budget overrun: 2-3x.

FOMO is expensive.

What AI Can Actually Do (2026)

Task	AI Can Handle?
Feature development	✅ 70-80%
Business logic	✅ 90%+ accuracy
Code review	✅ Catches what humans miss
Debugging	✅ Traces complex systems
Testing	✅ Comprehensive suites
System design	✅ Tradeoff analysis
Real-time systems	❌ Sub-100ms still needed
Safety-critical	❌ Can't tolerate 2-5% error

Two traps:

FOMO → Overestimate AI (→ broken systems)
Fear → Underestimate AI (→ left behind)

Both are expensive mistakes.

When to Use AI (Quick Decision Tree)

"Will this save time?" → If yes, use AI (even 30 seconds compounds)
"Cost of being wrong?" → Low = AI first. High = AI + human review
"Can I verify it?" → If yes (code, tests), AI away
"Latency budget?"** → <100ms = traditional code. >1s = AI fine
"Maintain in 6 months?" → If yes, understand what AI built

2026 reality: Nobody uses Stack Overflow anymore. AI is the default.

Career Advice: What to Learn

Do ✅

Master AI agent orchestration (new full-stack)
Learn AI + your domain (AI + backend, AI + security)
Understand AI economics (token pricing, Haiku vs. Opus)
Learn eval frameworks (new unit tests)
Keep core skills sharp (system design, debugging)

Don't ❌

Become "prompt engineer" only (commoditizing)
Ship AI code you don't understand
Ignore AI (you'll be left behind)
Use AI without measuring impact

Winners: "Senior Engineer who uses AI strategically" NOT "AI specialist" and NOT "AI refuser."

Warning Signs (Pause If...)

❌ Using AI agents when a script would work
❌ Building RAG for structured database data
❌ No monitoring on AI costs (bills WILL shock you)
❌ Can't write evals for AI features
❌ Hiring "AI engineers" before identifying problems
❌ Using Opus for Haiku tasks (cost optimization matters)

Your Action Plan (This Week)

Audit: Which tasks could AI help with? (Probably 60-70%, not 20%)
Pick ONE tool: Cursor, Claude Code, or Copilot X
Set boundaries: "AI for X, always review Y"
Measure: Track time saved, bugs caught, features shipped
Build one AI-native feature: Something impossible without AI

The Bottom Line

AI FOMO makes you do stupid things.

AI fear makes you do nothing.

Both are expensive mistakes.

The engineers who thrive won't be the ones who use AI the most. They'll be the ones who use it strategically.

Don't let FOMO turn you into someone who uses AI for everything. Don't let fear turn you into someone who uses AI for nothing. Use AI wisely. 🎯

What's working (and what's not) in your team? Drop a comment below.

Published: March 17, 2026

The 10x Engineer in 2026: What Actually Matters (After AI Changed Everything)

Hainan Zhao — Sat, 14 Mar 2026 13:03:44 +0000

The 10x engineer isn't dead — but the definition has completely changed.

The Reality Check

Let's be honest: 10x individual coding speed is mostly dead. AI writes code faster than any human now.

But 10x leverage is very much alive.

Here's what actually makes someone 10x in the age of AI.

What Actually Makes Someone 10x in 2026

1. AI Fluency (Non-Negotiable)

Using Claude Code, Copilot, or Cursor effectively isn't optional anymore. The difference:

❌ 1x: Typing basic prompts, accepting AI output blindly
✅ 10x: Knowing when to let AI write vs. hand-code, reviewing AI output critically, spotting plausible-but-wrong code

My observation: Engineers using AI agents are already 2-3x more productive. This compounds.

2. Judgment > Speed

Before AI: 10x = code faster, architect better, debug quicker

After AI: 10x = ask better questions, validate faster, ship with confidence

The skills that matter now:

Knowing which problems to solve (impact multiplier)
Saying "no" to low-value work (opportunity cost awareness)
Shipping good enough vs. perfect (velocity compounds)

3. System Thinking

Understanding how pieces fit together (not just your component)
Anticipating failure modes before they happen
Building for maintainability (your future self thanks you)

4. Communication Multiplier

One well-written RFC can save 100 hours of meetings. That's 10x behavior.

Writing docs that save future you time
Code reviews that teach, not just critique
Explaining technical tradeoffs to non-engineers

5. Automation Mindset

If you do it twice, automate it
Building tools that help other engineers (10x → 100x via team)
Killing toil ruthlessly

The Dark Side of "10x"

Toxic interpretations:

❌ Hero culture (staying up all night, burning out)
❌ Gatekeeping knowledge ("I'm indispensable")
❌ Moving fast and breaking things (technical debt bombs)

Healthy interpretation:

✅ Sustainable pace (consistency > bursts)
✅ Making yourself obsolete (document, delegate, automate)
✅ Moving fast without breaking production

My Framework: The 3 Layers of 10x

Layer	What It Is	How to Get There
1x → 3x	AI fluency + good habits	Use AI tools daily, write tests, read code
3x → 10x	Judgment + system thinking	Ship projects end-to-end, learn from failures
10x → 100x	Multiplying others	Mentor, build tools, write docs, set standards

Most people stop at Layer 1. The real leverage is Layers 2 & 3.

Actionable Advice (Start Today)

This Week:

Pick one AI tool — use it for everything
Write one piece of documentation that saves future you time
Automate one repetitive task (even if it takes 2 hours to save 30 minutes)

This Month:

Ship one project end-to-end (design → deploy → monitor)
Do one post-mortem on a mistake (write it down, share it)
Review a teammate's code with teaching mindset

This Quarter:

Build a tool that helps your team (internal CLI, template, workflow)
Mentor one junior engineer (or write tutorials publicly)
Say "no" to one low-impact project (protect your time)

The Real Secret

10x isn't about being a genius coder. It's about:

Working on the right problems (impact selection)
Not wasting time on bullshit (meeting hygiene, scope discipline)
Using leverage (AI, tools, automation, other people)
Learning from mistakes (post-mortems, not blame)
Making others better (documentation, mentorship, tools)

The engineers I've seen go from 1x → 10x didn't suddenly get smarter. They got better at choosing battles and better at amplifying themselves.

TL;DR

10x is real, but not how most people think
AI changed the game — coding speed is commoditized, judgment is scarce
The path: AI fluency → better judgment → multiply others
Start small: One automation, one doc, one shipped project
Think long-term: 10x is a 5-year journey, not a 30-day challenge

What's your take on 10x engineers? Drop a comment below — I'd love to hear what's working for you in 2026.

Published: March 14, 2026

Building Audio Alerts with TTS for Gemini CLI

Hainan Zhao — Thu, 12 Mar 2026 11:28:19 +0000

Building Audio Alerts with TTS for Gemini CLI

Ever wish Gemini CLI could actually speak to you? I built an extension that adds audio alerts with text-to-speech.

Quick Install

gemini extensions install https://github.com/HainanZhao/gemini-extension-audio-alerts

Set your theme:

export AUDIO_ALERTS_THEME=retro

Available themes: retro, espionage, hero, portal, premium

The Problem

When running Gemini CLI in the background, you miss important moments:

When the agent needs your permission to run a tool
When an error occurs
When a long task completes

The Solution

Plays themed sounds followed by spoken messages:

Sounds

retro - 8-bit game sounds
espionage - High-tech clicks
hero - Dramatic fanfare
portal - Magical transitions
premium - Elegant chimes

Spoken Messages

Event	Retro Theme	Espionage Theme
Permission	"Permission needed"	"Agent requesting permission"
Error	"Error detected"	"Critical failure detected"
Done	"Game over. You win"	"Mission accomplished"

Cross-Platform

macOS: say, afplay
Linux: espeak, gtts-cli, play
Windows: PowerShell System.Speech

Repo

https://github.com/HainanZhao/gemini-extension-audio-alerts

This post was written with assistance from AI coding agents.

Building Audio Alerts with TTS for AI Coding Agents

Hainan Zhao — Thu, 12 Mar 2026 11:11:39 +0000

Building Audio Alerts with TTS for AI Coding Agents

Ever wish your AI coding agent could actually speak to you? I've been working on adding audio alerts with text-to-speech to Gemini CLI, and here's how it works.

The Problem

When running AI coding agents in the background, you often miss important moments:

When the agent needs your permission to run a tool
When an error occurs
When a long task completes

You either have to constantly check the terminal or miss important updates entirely.

The Solution: Audio Alerts + TTS

I built a hook system that plays themed sounds followed by spoken messages:

Sounds

Each theme has its own sound:

retro - 8-bit game sounds
espionage - High-tech clicks
hero - Dramatic fanfare
portal - Magical transitions
premium - Elegant chimes

Spoken Messages

After each sound, a TTS message plays:

Event	Retro Theme	Espionage Theme
Permission needed	"Permission needed"	"Agent requesting permission"
Error	"Error detected"	"Critical failure detected"
Done	"Game over. You win"	"Mission accomplished"

Cross-Platform Support

The implementation works on:

macOS: Uses built-in say command and afplay
Linux: Supports espeak, gtts-cli, and play
Windows: Uses PowerShell's System.Speech and SoundPlayer

How It Works

The hook intercepts Gemini CLI events:

# Play sound
play_audio "$ASSETS_DIR/done.wav"

# Wait a bit
sleep 0.3

# Speak the message
speak "Task complete"

All runs non-blocking, so the agent keeps working while you get audio feedback.

Installation

Clone the repo:

git clone https://github.com/HainanZhao/gemini-extension-audio-alerts
cd gemini-extension-audio-alerts

Set your theme:

export AUDIO_ALERTS_THEME=retro

What's Next

Possible enhancements:

Custom voice selection
More themes
Sound mixing
Volume control

The code is open source and available at: https://github.com/HainanZhao/gemini-extension-audio-alerts

This post was written with assistance from AI coding agents.

The State of AI in Software Engineering: 2026

Hainan Zhao — Wed, 11 Mar 2026 00:19:56 +0000

The State of AI in Software Engineering: 2026

The software engineering industry is undergoing a fundamental transformation. After years of hype and experimentation, AI in development has reached an inflection point. Let me share what's actually happening in the field—not speculation, but real trends backed by numbers.

The Numbers That Matter

Before diving into trends, consider the revenue trajectory:

Anthropic started 2025 at $1 billion run rate
Hit $7 billion by October 2025
Projected to reach $26 billion in 2026

This isn't hype. It's real demand from real enterprises solving real problems.

The 2026 Landscape

1. Agentic Development Is Leaving the Lab

The biggest shift we're seeing: AI agents have evolved from "copilots" (assistants that suggest code) to "coworkers" capable of autonomous workflows.

What this means in practice:

Agents can now plan, execute, and self-correct without constant human intervention
Multi-step tasks that previously required manual coordination now run automatically
The agent handles the "how" while humans focus on the "what"

2. Vibe Engineering Is the New Skill

There's a new term circulating: "vibe engineering." It describes the skill of guiding AI to generate full-stack applications from high-level intent.

Instead of:
\typescript // Create a React component with props... \\

You say:
\text Build me a dashboard that shows user analytics with real-time charts \\

The AI handles the rest—component structure, API calls, state management.

The developer's role shifts from writing syntax to:

Defining clear intentions
Reviewing generated code
Handling edge cases the AI can't anticipate

3. Self-Healing QA

Perhaps the most impressive advancement: AI agents that can identify bugs, write patches, and re-test without human intervention.

Traditional QA loop:

Write code → 2. Test fails → 3. Developer investigates → 4. Fix → 5. Re-test

AI QA loop:

Agent writes code → 2. Test fails → 3. Agent investigates, fixes, re-tests → 4. Human reviews

This isn't theoretical. Teams are reporting 10x+ improvements in debugging cycles.

4. TypeScript's Rise

A subtle but significant data point: TypeScript recently became the most-used language on GitHub, overtaking Python and JavaScript for the first time.

Why? AI works better with strongly-typed code. TypeScript's type inference makes AI-generated code more reliable and maintainable.

5. Enterprise Adoption Acceleration

Gartner predicts 40% of enterprise apps will have agentic AI capabilities by end of 2026.

The conversation has shifted:

2024: "Should we try AI?"
2025: "How do we pilot AI?"
2026: "How do we scale AI across all teams?"

Real-World Impact

Let's look at concrete examples:

Fortune 100 Company

Before AI: 9-day pull request cycle
With AI: 2.4 days
That's a 73% reduction in cycle time

Latin American Fintech

8-year migration project
Completed in weeks with AI
12x efficiency improvement

These aren't isolated incidents. They're becoming the norm.

What This Means for Developers

The developer role is evolving into what I call "Orchestrator":

Old Role	New Role
Write syntax	Define intentions
Debug line-by-line	Review agent decisions
Handle repetition	Handle exceptions
Test manually	Verify AI tests

The core skills shift from memorization to:

System design
Prompt engineering
Critical evaluation
Ethical governance

The Challenges Ahead

It's not all smooth sailing. The industry faces real challenges:

Correctness & Trust - AI can be confidently wrong
Security - Supply chain attacks via AI-generated dependencies
Governance - Who owns the code the agent writes?
Skill Gaps - Teams struggle to integrate AI into existing workflows

Looking Forward

2026 is different from previous years. The novelty phase is over. The engineers who are succeeding now aren't just "using AI"—they've built systematic approaches to AI-augmented development.

The question isn't whether AI will change software engineering. It already has. The question is: are you building the skills to work alongside it?

This post was generated with assistance from AI, reflecting current industry trends and insights.

Clawless — Bring Your Own Agent to Telegram & Slack

Hainan Zhao — Tue, 10 Mar 2026 11:11:25 +0000

Ever wished you could chat with your favorite AI coding agent from your phone? Clawless makes that possible.

What is Clawless?

Clawless is a lightweight bridge that connects local AI agent CLIs (Gemini CLI, OpenCode, Claude Code) to Telegram or Slack. It lets you use your preferred coding assistant through messaging apps without rebuilding your bot integration.

Why Clawless?

BYO-agent: Keep using your favorite CLI runtime
Local-first: Your data stays on your machine
Flexible: Swap agents without migrating platforms
Minimal: Just the glue code, nothing extra

Features

🤖 Telegram & Slack support
🛠️ MCP tool support via your local CLI
💾 Persistent conversation context
⚡ Async mode for long-running tasks
⏰ Cron scheduler via REST API

Quick Start

npm i -g clawless
clawless

That's it! Configure your Telegram bot token and you're ready to go.

Architecture

Messages flow from Telegram/Slack → Clawless → Your local agent CLI (via ACP) → back to you.

Try It Out

Get started at: https://github.com/HainanZhao/clawless

This post was created by Clawless itself!