DEV Community: Okkar Kyaw

Vibe coding won’t make your product stand out. Here’s what will

Okkar Kyaw — Sat, 17 Jan 2026 22:32:54 +0000

You're paying for AI. $20/month, maybe $200. Of course you want it to be magic.

That's the trap. You paid money, so your brain wants results. You want to describe what you want and have it appear, perfect, ready to ship. Like a genie.

But here's what actually happens: you slowly accept "good enough." You keep prompting, keep iterating, keep spending tokens. Before you know it, your product looks like everyone else's product. You became a boiling frog and didn't notice.

The ones who stand out in 2026 accepted a different reality. AI is leverage, not a magic pill.

The slopidemic

"AI Slop" was Word of the Year 2025. Merriam-Webster and the American Dialect Society both picked it.

Everyone has access to the same tools. v0, lovable, bolt, cursor, gemini, claude, you name it. All producing similar outputs when used the default way. 82% of AI-generated sites/apps share the same layouts. Same onboarding flows. Same card-based UIs. Same everything.

When your product looks like everyone else's, you're invisible. 1.2% conversion rate for generic AI output vs 2.8% for custom work. That's 2.3x worse. Brand recall drops 25% when there's nothing distinctive to remember.

55% of consumers feel uncomfortable when they sense AI was behind the marketing. Not because AI is bad. Because they can tell nobody cared enough to make it unique.

If you're shipping slop, you're cooked.

The hierarchy that works

Stop expecting AI to be the genie. Start with specialized solutions first.

1. Expert, agency, or team
Hire someone who understands both design and development. Someone who knows what distinctive looks like in your market. Still the gold standard.

2. Specialized platforms
Find tools built by people who understand your specific craft. Not generic AI builders, but platforms where the craft comes first and AI assists.

My picks: Framer for websites, Figma for design. Claude for coding. Find yours for your domain.

3. AI as leverage on top
Now add AI. Not as the genie that does everything. As the multiplier that accelerates what you're already doing.

Most people reverse this. They start with generic AI tools, hope for magic, get slop.

The question isn't whether to use AI. It's 2026. Everyone uses AI. The question is: are you using it to do the same thing as everyone else, or to multiply your edge?

Case Study

Maryam is an Apple software engineer building a matcha community called MBM (Matcha by Maryam). Events, content, products, partnerships across the US and Canada.

She had a Next.js site she built "just to share" her matcha journey. Functional but generic. Looked like any other hobby blog. No time to make it special. Work at Apple, sourcing matcha from Japan, planning events. Something had to give.

Problem: when you're pitching to brands, your website IS your credibility. Generic website means generic perception.

She could have kept prompting AI tools. Kept hoping for the genie. That's the boiling frog trap.

What we did instead: specialized solution + AI leverage.

First, strategy questions. What should visitors feel? What makes MBM different from other matcha content? What patterns are competitors overdoing?

Then Figma/Gemini prototyping. Built with Framer. AI for speed and iteration. Human direction for differentiation.

The before: generic site, could have been any matcha blog.

The after:

Instant welcoming and memorable EXPERIENCE for target audience
The right green that felt like nature, growth, freshness.
Minimal motion. Enough to feel alive and premium, not gimmicky.
Copy speaking directly to matcha lovers. Inside jokes. The feeling of belonging.

Result: people praise her. They trust her more to do deals. The website became an asset, not just a checkbox. +65% trust boost with human refinement is real.

That's what happens when you stop hoping for the genie and start using the hierarchy.

P.S. We also extended to brochures, Luma-inspired app designs, React Native prototypes. AI accelerated all of it. But the manual craft came first.

The multiplier

Most people use AI to reduce effort. Same output, less work.

Flip it.

Use AI to multiply output. Same effort, more results.

SPEED × QUALITY × VOLUME

Not additive. Multiplicative.

10x iterations. 10x experiments. 10x directions explored. That's where distinctive comes from. Not from one magic prompt, but from the volume of attempts with human direction.

Closing

Accept reality. Think different. Act different.

When everyone uses the same tools the same way, the ones who stand out are the ones who refused to settle for the default output.

MCP Was a Context Killer. Claude Code Just Fixed It.

Okkar Kyaw — Fri, 16 Jan 2026 01:29:33 +0000

One of the biggest problems right now for everyone using AI coding tools is context window and MCP.

We only have 200K context window. And we don't even get full allocation for that when we code. Between system prompts, conversation history, and tool definitions, you're already starting at a deficit.

Well, Claude Code just released MCP Tool Search.

The Context Tax

The biggest issue with MCPs is they're powerful but expensive.

Just 5-6 MCP integrations will fill up your context window and you can't even do anything. You connect Figma, Playwright, GitHub, a couple custom servers, and suddenly 30-50% of your context is gone before you type a single prompt.

Opus is powerful. But what's the point of using it when you're running out of session every 10-15 minutes?

We can't ignore MCPs either since it's the easiest and most popular way to extend Claude's capabilities. Browser automation, database access, file operations, custom APIs. This is what makes Claude Code actually useful for real work.

So we were stuck. Use MCPs and burn context. Or skip MCPs and lose functionality.

What We Tried

A couple months ago, solutions started coming out. Anthropic, Cloudflare, Docker all mentioned their suggestions and approaches. People tried them, including me.

Code execution wrappers
Skills with lazy loading
Universal MCP configs
Manual tool prioritization

They all have pros and cons. Some broke tool discovery. Others added latency. None felt like real solutions.

I've been waiting for an official fix. And now it's here.

The Fix

It's awesome and intuitive.

Connect whatever MCP you want and Claude will search and load relevant tools for you. Instead of preloading every single tool definition at session start, it searches on-demand.

Here's how it works:

1. Threshold detection: If your MCP tools exceed 10% of context, Tool Search activates automatically.

2. Semantic search: When you need a tool, Claude searches for it by keyword and similarity. Need GitHub tools? It searches "github" and loads just those definitions.

3. On-demand loading: Tools only consume context when you actually use them. Switch tasks? Context shifts with you.

No config needed. It just works.

The Proof

Before: MCP tools eating 20-50% of context window

After: MCP tools loaded on-demand, effectively 0% until needed

Simon Willison said it best: "Context pollution is why I rarely used MCP, now there's no reason not to hook up dozens or hundreds of MCPs."

That's exactly right.

If It's Not Working Yet

Since it's rolling out, it might not be available for you even though your CLI is up to date.

For weird reasons, it happened to me. If it's happening to you, do this:

source ~/.zshrc
export ENABLE_TOOL_SEARCH=true
claude

That's it.

Check with /context. If MCP tools shows "loaded on-demand" instead of a token count, it's working.

Call some MCP tool and you'll see MCPSearch appear when Claude finds the right integration.

Go Wild

Now I'm enabling all the MCPs I've been waiting to use. I'm Lovin' It.

Notion, Linear, Exa, Vercel, database tools, the whole stack. There's no penalty anymore.

The barrier that kept most people at 2-4 servers is gone. Connect everything. Let Claude figure out what it needs.

Time for you to get wild too. And let's find something else to complain to Anthropic about lol

Google's Universal Commerce Protocol: What Developers Need to Know

Okkar Kyaw — Mon, 12 Jan 2026 03:33:42 +0000

Google doesn't want YOU to shop anymore.

They just announced the Universal Commerce Protocol, and if you're building anything in e-commerce, payments, or AI agents, this changes your roadmap.

What Is UCP?

UCP is a universal translator for AI shopping. Imagine telling your phone "Find me a carry-on bag under $200" and instead of just searching, the AI compares prices, checks reviews, and actually buys it for you. No checkout forms. No clicking through five websites. The AI handles everything.

The key technical detail: UCP is an open standard that unifies discovery, cart management, checkout, fulfillment, and post-purchase support across AI platforms and retail backends.

UCP powers native checkout in Google's AI Mode within Search and the Gemini app, starting with Google Pay. Retailers stay in control as the seller of record.

For the technical deep dive, check out Google's Developer Blog on UCP internals.

The Numbers That Matter

Early pilot tests tell an interesting story.

Metric	Data	Source
Consumer trust in AI purchases	24%	Forrester
Sales increase in pilot tests	122%	AInvest/NRF
US agentic commerce by 2030	$190-385 billion	Morgan Stanley
Global market by 2030	Up to $1 trillion	McKinsey
Want control over final decision	60%	BCG/LinkedIn

Here's the tension: only 24% of consumers currently trust AI to make purchases for them. Google is betting that convenience, security, and standardization will change that number.

Who's In and Who's Out

Google co-developed UCP with major players.

Core Partners:

Shopify, Walmart, Target, Etsy, Wayfair

Payment Processors:

Mastercard, Visa, Adyen, Stripe, American Express, Google Pay, PayPal

Notably Absent:

Amazon (building their own "Shop Direct" instead)
Apple (data sovereignty concerns)

Amazon's absence is telling. The biggest e-commerce company in the world is reportedly building their own agentic API rather than joining Google's consortium. TechCrunch has the details on why Amazon and Apple are sitting this out. This sets up a potential protocol war.

The Protocol Landscape

If you're building AI agents, you need to understand where UCP fits.

Protocol	Owner	Focus	Limitation
UCP	Google + partners	Full journey: discovery to fulfillment	New, adoption uncertain
MCP	Anthropic	Context exchange between AI models	Limited payment primitives
A2A	Open standard	Peer-to-peer agent communication	No unified commerce lifecycle
ACP	OpenAI/Stripe	Checkout and payment flows	Closed-source components

UCP's advantage: it covers the full commerce lifecycle. MCP handles context. A2A handles agent communication. UCP handles buying stuff.

For a detailed comparison, see Auth0's Protocol Comparison Guide and OneReach's MCP vs A2A analysis.

What This Means for Developers

Three scenarios to consider:

If you're building AI agents: UCP gives you a standardized way to interact with retail backends. Instead of building custom integrations for each merchant, you implement UCP once.

If you're building for e-commerce: You're facing a choice. Integrate with UCP to be discoverable through Google's AI Mode and Gemini, or wait and see if Amazon's approach wins.

If you're building payments infrastructure: The protocol war is happening now. MCP, A2A, ACP, and UCP are competing for the standard layer. Betting on one means betting against others.

Check out Google Cloud's implementation guide for getting started with UCP.

The Privacy Question

This isn't all upside. Privacy advocates have raised concerns.

Digital Rights Watch warns that UCP could centralize transaction data within Big Tech infrastructure. The Electronic Frontier Foundation is calling for explicit, auditable consent protocols and on-device data handling.

When one company can see what you search for, what you browse, and what you buy, the data aggregation creates unprecedented visibility into consumer behavior.

Google emphasizes that retailers remain the seller of record. But the infrastructure still flows through Google's systems.

Small Business Reality

Small merchants report anticipated onboarding costs of $15,000 to $25,000 for UCP compliance. This includes developer resources, API modernization, and staff training.

According to a December 2025 survey by the National Federation of Independent Business, 38% of small retailers express skepticism about AI-driven commerce standards.

On the flip side, UCP could help small businesses compete. A boutique could suddenly be discoverable through Google's AI Mode, appearing alongside major retailers in AI recommendations.

What to Watch

Several dynamics will shape how this evolves:

Consumer adoption: Will people actually let AI buy things for them?
Amazon's response: If Shop Direct gains traction, merchants may face ecosystem choices
Regulatory attention: Given ongoing antitrust scrutiny of Google, regulators may examine UCP's market implications
Developer tooling: The quality of SDKs and documentation will determine adoption speed

The Uncomfortable Truth About Open Source in the AI Era

Okkar Kyaw — Fri, 09 Jan 2026 20:29:48 +0000

The most popular CSS framework just fired 75% of their engineering team.

Downloads up 5x. Revenue down 80%. Docs traffic down 40%.

This isn't a failure story. It's worse. It's a success story that's killing them.

You're Probably Using Tailwind Right Now

If you've ever vibe coded anything... Cursor, v0, Claude Artifacts... you're using Tailwind and you might not even know it.

It's the CSS framework that powers basically everything AI builds. When you ask Claude to make a landing page, that's Tailwind. When Cursor autocompletes your styles, that's Tailwind. When v0 generates a component, that's Tailwind.

Metric	2023	2026	Change
npm Downloads	6M/week	32M/week	+433%
Revenue	Baseline	-	-80%
Docs Traffic	Baseline	-	-40%
Engineering Team	4	1	-75%

AI learned Tailwind so well that nobody needs to learn it anymore.

How Open Source Actually Makes Money

Here's what most people don't understand about open source business models.

Tailwind CSS is free. Always has been. The way they pay their team is through Tailwind Plus... beautifully designed templates and UI kits. You buy them once, you use them forever.

The only way people find out about these paid products is through the documentation. That's their entire funnel. Developers hit a problem, visit the docs, see the premium offerings, some percentage converts.

But here's the thing.

AI doesn't visit docs.

You don't visit docs anymore either. When's the last time you actually went to tailwindcss.com instead of just asking ChatGPT?

And now you've got shadcn/ui offering similar components for free. Who's buying templates when Claude can generate a decent hero section in three seconds?

"The Tailwind team is like musicians... they go into a hole, work on an album, do a drop, go on tour, and that's where they make the money. But you miss out on all the advantages of compounding." - Dax from SST

A customer you grabbed five years ago should still be paying you. That's the problem with one-time purchases in a world where AI makes everything feel disposable.

// Detect dark theme var iframe = document.getElementById('tweet-2009068367005581555-181'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2009068367005581555&theme=dark" }

Adam Told the Truth

Then Adam Wathan did something most founders never do. He told the truth.

He posted a 33-minute voice recording while walking through the snow in Ontario. Just talking to his phone. Raw. Unedited.

And what he said was brutal.

He called it a "boiling frog" situation. Revenue had been declining so slowly for so long that he didn't even notice until he finally ran the numbers over the holidays.

The forecast showed that if nothing changed, in about six months they wouldn't be able to make payroll. So he made the call. Laid off three of his four engineers NOW so he could give them proper severance instead of waiting until the money ran out completely.

Then he wrote this on GitHub:

"75% of our engineering team lost their jobs yesterday because of the brutal impact AI has had on our business."

"I feel like a FAILURE for building this insanely successful open source thing where business success is inversely correlated with the project's popularity."

Read that again. The more successful Tailwind becomes, the faster their business dies.

// Detect dark theme var iframe = document.getElementById('tweet-2008909129591443925-537'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2008909129591443925&theme=dark" }

This Isn't Just Tailwind

Vjeux, the guy behind React Native, Prettier, and Excalidraw, said he has the exact same problem. Prettier is used by basically everyone in the industry, but funding proper maintenance has been a constant struggle.

Jeffrey from Laracasts had to do layoffs the same week.
// Detect dark theme var iframe = document.getElementById('tweet-2009052644317479016-103'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2009052644317479016&theme=dark" }

badlogicgames tweeted something that got over a million views: "Everything is going to be f**ked real soon."

// Detect dark theme var iframe = document.getElementById('tweet-2008957430441373811-119'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2008957430441373811&theme=dark" }

Here's the pattern.

AI companies are making billions. The open source projects they trained on? The ones their products literally can't function without? They're dying.

Not because they're failing. Because they're too successful.

Tailwind is the first domino. It won't be the last.

There's Actually Hope

Since this news broke, donations have been pouring in. Adam posted a thank you tweet that got over a hundred thousand views. The community is rallying. People are actually signing up for Tailwind Plus, buying the Refactoring UI book, becoming sponsors.

The response has been overwhelming. Developers are waking up to the fact that the tools they rely on are built by people who need to eat.

// Detect dark theme var iframe = document.getElementById('tweet-2009017727592353959-544'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2009017727592353959&theme=dark" }

What You Can Do

If you use Tailwind... and statistically, you probably do... here's what you can do right now:

1. Buy something. Refactoring UI is genuinely one of the best design books for developers. Tailwind Plus has templates that will save you hours.

2. Sponsor on GitHub. Even $5/month adds up when thousands of developers do it.

3. Check your other dependencies. Run npm ls and look at what you're using. See who else is struggling to keep the lights on.

4. Share this story. Most developers have no idea this is happening. The more people who understand the problem, the better chance we have of fixing it.

Because if we don't support these projects, there won't be a Tailwind v5. There won't be a Prettier update. There won't be any of the tools we rely on every single day.

The tools you rely on are built by people who need to eat. That's it. That's the whole thing.

Let's Talk About This

I want to hear from you:

What open source project do you depend on that might be struggling right now?

Drop it in the comments. Let's surface the projects that need support and actually do something about it.

Follow me for more AI and tech breakdowns. For deeper analysis, subscribe to Oak Codex Newsletter for FREE.

From ‘Shell Product’ to $2 Billion: The Manus AI Story Nobody Saw Coming

Okkar Kyaw — Wed, 07 Jan 2026 02:56:36 +0000

Nine months. That's all it took for Manus AI to go from being called an "unusable shell product" to closing a $2 billion acquisition with Meta.

This isn't just an acquisition story. It's a signal about where AI is heading, and what it means if you're building products in this space.

The Demo That Broke the Internet

March 2025. Manus drops a demo that immediately goes viral. The pitch was simple but revolutionary: this is the first "general AI agent," an AI that actually does things, not just talks about them.

You give it a task. You watch it plan. You watch it execute. You get real results.

No wall of text. No "here's how you could do it." Actual output.

"The AI that DOES" became Manus's viral tagline, and it hit different. This wasn't another chatbot. This was something new.

The hype was immediate and intense. Invite codes were selling for $14,000 on Xianyu, China's secondary marketplace. People were calling it the next DeepSeek moment. The transparency layer, where you could literally watch every action the agent takes in real time, felt genuinely innovative.

But here's the thing about hype. It's a loan you have to pay back with results.

And Manus was about to default.

The Brutal Backlash

Chinese social media turned hostile fast. The accusations were brutal and specific:

"Pure shell product": critics claimed it was just a wrapper on Claude with nothing original
"Hunger marketing": accusations that they raised money and ran
800 credits to upload a single file, users were burning through resources on basic tasks
Reddit threads calling it "absolutely unusable"

Most startups die at this stage. Once the "scam" narrative takes hold, you can't outmarket your way out of it. The product had real problems, and the court of public opinion had already rendered its verdict.

When your product is called a "shell product" by the very market you're trying to capture, that's usually a death sentence.

But Manus didn't die. They made a decision that would change everything.

The $100 Million Resurrection

Instead of chasing Twitter virality, Manus started chasing invoices. They went completely heads down on product.

June 2025: Playbook templates launched, pre built task flows that cut credit consumption by 30%. Users could now accomplish complex workflows without burning through their allocation.

July 2025: Chat Mode introduced, simple queries now used 5 credits instead of 100. The product became actually usable for everyday tasks.

Then the enterprise deals started landing. Mercury Law, a UK firm, started using Manus for contract review and reported 40% time savings. Shanghai Zenith Advertising adopted it for campaign automation. Beijing consultancies integrated it into research workflows.

Real companies. Real invoices. Real validation.

By December 2025: $100 million ARR. Two million users.

From "shell product" to nine figure revenue in six months. That's not a pivot. That's a resurrection.

Why Meta Paid $2 Billion in 10 Days

Normal M&A takes months of due diligence. Legal review. Integration planning. Negotiation rounds.

Meta closed in 10 days.

That's not strategic M&A. That's desperation. And when you understand the competitive landscape, it makes perfect sense.

Here's what Meta was facing:

Google: Gemini Agents rolling out across Workspace
Microsoft: Copilot embedded in everything, everywhere
OpenAI: Function calling agents shipping rapidly
Anthropic: Claude becoming the go to for agentic workflows

Everyone was moving. Meta was stuck.

Meta didn't just buy revenue. They bought an agentic architecture they were years behind on building themselves.

And here's what nobody's talking about: Meta is quietly shifting away from open source. They're building a proprietary model codenamed "Avocado." The "Llama is free forever" era? That's changing. They need revenue from AI, not just developer goodwill.

According to Gartner, 40% of enterprise apps will have AI agents by end of 2026, up from 5% the year prior. Meta paid $2 billion to be on the right side of that shift.

What This Means for Developers

This acquisition isn't just a Manus story. It's a signal about where AI is heading, and what it means if you're building products today.

Chatbots chat. Agents act.

The difference is fundamental:

Chatbots	AI Agents
Respond to prompts	Execute multi step tasks
Generate text	Deliver outcomes
Require human orchestration	Operate autonomously
Useful for Q&A	Useful for work

The transparency layer that Manus pioneered, watching the AI think, plan, and execute in real time, is becoming table stakes. Users don't just want answers anymore. They want to see the work happening.

If you're building AI products, the question isn't "how do I make my chatbot smarter?" It's "how do I make my AI actually do things?"

That's the paradigm shift. And Big Tech is paying billions to catch up.

Key Takeaways

The Manus story offers several lessons for anyone watching, or building in, the AI industry:

Product beats narrative. All the viral marketing in the world couldn't save Manus when the product failed. Only fixing the product did. The criticism was valid. They addressed it. That's what mattered.

Enterprise revenue validates. $100M ARR in six months turned "shell product" criticism into a $2B exit. Revenue is the ultimate credibility signal.

Timing matters more than ever. Meta was behind on agents precisely when the market shifted from chatbots to agents. That timing created a $2B opportunity.

Speed signals desperation. 10 day M&A cycles don't happen unless someone is scared of being left behind. Meta was scared.

From shell product to $2 billion in 9 months. Failure to acquisition. Hate to exit.

That's the Manus story. And it's a preview of how fast the AI industry moves, and how quickly fortunes can reverse.

Let's Discuss 💬

What's your take on the chatbot to agent shift? Are you building anything with agentic AI? I'd love to hear your experience in the comments.

Follow me for more AI and tech breakdowns. For deeper analysis, subscribe to Oak Codex Newsletter for FREE.

Why Chatbots Are Dead in 2026 (And What's Replacing Them)

Okkar Kyaw — Sat, 03 Jan 2026 00:02:37 +0000

I'm going to make a prediction that sounds insane: by 2026, chatbots are officially dead.

Not the technology itself. The experience. We're going to look back at 2024 and 2025, all that time we spent typing paragraphs into a box and waiting for walls of text, and realize how absolutely broken that was. Because the future isn't AI that talks about doing things. The future is AI that actually does them.

And most of the apps you're using right now? They're not ready.

The New Definition of AI Slop

We've been calling apps with purple gradients and broken UI layouts "AI slop" all year. But here's a better definition.

99% of AI apps right now swear they're going to change your life. All that marketing about AI agents, AI technology, revolutionary workflows. Then you open the app and get the same three little dots, followed by a wall of text. A paragraph describing what you already know.

It doesn't do anything. It's a chatbot cosplaying as a tool.

Even when these apps do something, it takes forever or the actions are useless. They're not AI agents. They're the new AI slop.

The reason most AI apps feel the same is that they all copy each other. We have a term for this: wrappers. Apps whose entire "AI engine" is just a lazy prompt template hitting GPT or Gemini or Claude. No real workflow. No actual domain expertise. No ability to take useful action in your life.

50% of AI funding went to these wrapper companies. But we're in 2026 now. We can't just wrap AI models with some system prompts and call it a day.

Even the big companies are struggling. I love Perplexity's Comet Browser, but most of the time when it controls pages, it takes forever to figure out what I can just click in seconds. It lags manual browsing by 30% in multi-step tasks. We're still in early phases.

And Apple Intelligence? Remember that WWDC hype about Siri doing things for you? Craig Federighi admitted the features were "not ready" and delayed them to 2026. Even the companies that could build better are stuck in too much politics, too many committees, too many approval processes. That's where indie developers and startups have the strength and adaptability to actually experiment.

What Actually Works

Here's what separates the apps that work from the slop.

I typed one sentence into YouSoul: a MVP app i've been working on. The AI didn't give me a paragraph describing what I said. It created three actual tasks. And here's the key part: it scheduled them intelligently. Workout at 7am because exercise belongs in the morning. Blog post at 10am because deep work needs focus time. The AI even understood the vibe of each task, adding mood indicators automatically.

You talk to it like a human, and it takes action like an assistant.

Slop talks about doing things. Real apps actually do them. Use this as a filter for every AI app you try.

The apps that work share three characteristics:

They execute, not describe. When you ask to schedule something, it appears on your calendar. Not a paragraph about scheduling.

They understand context. No more "what date?" "what time?" "which contact?" The AI infers from context. "Morning workout" means 7am, not a philosophy lecture about exercise.

They feel like experiences. Sound design, motion, atmosphere. Not the existential dread of a gray form waiting for your input.

Remember, this is still the bare bones of what AI agents should be.

The Future: Generative UI

Even the best action-based apps are still the bare-bones of what agents should be. The real endgame is generative UI.

Google just announced this in their Research Blog. The AI doesn't just respond with text. It builds an entire interface custom for your question. Interactive tools, simulations, experiences, all generated on the fly.

Generative UI: A rich, custom, visual interactive user experience for any prompt

research.google

There's already movement in this direction. GPT Stores offer some of this functionality. Vercel AI SDK provides generated UI components. Dedicated solutions like C1 Thesys are emerging. But right now they're way too expensive, charged on top of the base AI models.

We're not fully there yet. But this is the direction. If your favorite AI app isn't thinking about generative interfaces, they're already behind.

The Developer-Consumer Gap

Here's what's actually wild about the current state of AI.

AI agents are incredible right now if you're a developer. Cursor, Claude Code, coding agents in general are legitimately shipping production code. But for everyone else using consumer apps? We're still in early phases. Consumer chatbots and agents are stuck at 60% accuracy.

The gap is real. And I think by 2026, that gap closes violently. Or users stop tolerating it entirely.

Apps With Soul

Even when apps work, they're still boring. They're utilities. Click, type, submit, wait. The UI might look fine, but there's no response. There's no atmosphere.

Take a look at gaming. Games have vibes. Sound design, motion, atmosphere. Every interaction feels intentional.

For software development and applications, that's traditionally not the focus. But with the power of AI, we can push much more on quality than on quantity. We can build apps that don't just function, they feel.

The Progression

In 2024, we were all impressed that computers could talk to us. Whole apps were built around that single novelty. Type something, get text back, wow.

In 2025, we were told "we're in the age of AI agents." But talking isn't enough. Text responses and bare-bone simple agent tool calls don't actually change your life. They just describe how your life could be changed.

By 2026, the shift isn't just about companies delivering. It's about expectations rising. Gartner predicts AI agent spend will hit $300 billion by 2026. Meanwhile, chatbot-only solutions are contracting 5% annually. If your AI doesn't do things, real actions, real scheduling, real task creation, it basically doesn't exist. Users won't tolerate chatbots anymore.

The winners won't be the chatboxes. The winners will be the agents that act and deliver. The apps that feel like experiences, not utilities. The tools that understand your life and actually participate in it.

A lot of big companies will get stuck in meetings. The indies? Shipping.

Stop settling for AI slop. Demand tools that work. And if you're building something, build something with soul.

Follow me for more thoughts on AI that actually works, not just talks about working.

What's your experience with AI apps? Are they actually helping or just generating text? Drop a comment below. 👇