What's Coming in AI (And How to Prepare)

Eight weeks ago, we started with a simple question: what is an AI coding agent?

Now you know. You've set one up. You've picked a model. You've built a landing page. You've seen the benchmarks, challenged the prompt engineering hype, looked inside a real business workflow, and assembled a complete free tool stack.

That's a lot of ground in eight weeks. But here's the thing: the ground is still moving. The tools you set up this month will be different in six months. The models you benchmarked will be replaced. The "impossible" things will become checkboxes.

So let's talk about where this is going. Not in vague visionary terms. In specific, practical terms based on what's already happening.

---

The Pace of Change

First, some perspective. Here are things that were hard, expensive, or impossible six months ago that are now trivial:

• Building a full web application from a description. In late 2025, AI coding agents could write individual functions. Now they build complete applications from a paragraph of English. You did this yourself in Week 4.

• Accessing multiple models through one interface. You used to need separate accounts for GPT-4, Claude, and Gemini. Now OpenRouter gives you all of them through a single API key. The switching cost between models went from "create account, add payment, learn new API" to "change a dropdown."

• Deploying for free. Cloudflare Pages offered free hosting before, but the ecosystem around it (automatic deploys, edge functions, free SSL) has made the paid tier irrelevant for most small projects.

• Image generation quality. Six months ago, free image generators produced obviously AI-looking images. Now the free options produce images that are good enough for blog headers and social posts.

This pace isn't slowing down. If anything, it's accelerating. And that creates both an opportunity and a problem.

---

What's Coming in the Next 6-12 Months

Based on what's already in development, in beta, or shipping in limited release, here's what I expect to change:

Agents That Run Autonomously

Right now, AI coding agents respond to your prompts. You say "build this," it builds it. You say "fix that," it fixes it. You're in the loop for every action.

That's changing. The next generation of agents will run autonomously -- you give them a high-level goal ("add user authentication to my app and make sure it passes security best practices"), and they plan, execute, test, and iterate without needing your input at every step.

This is both exciting and dangerous. Exciting because you'll be able to delegate entire features. Dangerous because "just let the AI do it" without understanding what it did is a recipe for bugs, security holes, and technical debt.

The skill you'll need: knowing when to delegate and when to stay in the loop. The more you understand about what the agent is doing, the better you'll be at this judgment call.

Multi-Modal Coding

Right now, you describe what you want in text. Soon, you'll be able to show the agent a screenshot of a design and say "make this." Or point your camera at a whiteboard sketch and say "build an app that looks like this."

This removes the last major barrier between "I know what I want" and "I can describe it in words." For visual thinkers, this is huge.

AI-Native Applications

We're going to start seeing products that only exist because AI makes them possible. Not "existing product with AI features bolted on" -- entirely new categories of software that couldn't exist without AI.

Think: apps that generate themselves based on your specific needs. Tools that adapt their interface to your skill level. Services that create custom workflows by watching how you work.

The Gap Widening

Here's the one that matters most: the gap between people who can use AI tools and people who can't is going to widen. Fast.

It's not about being "into AI." It's about productivity. Someone using AI tools effectively can produce in hours what takes someone else days. In a competitive environment -- job market, freelance market, startup landscape -- that advantage compounds.

The people who started eight weeks ago and built something real? They're ahead. Not because they're smarter. Because they started.

---

How to Prepare Without Getting Overwhelmed

The biggest risk isn't falling behind. It's getting overwhelmed and doing nothing. Here's how to avoid that:

1. Pick ONE tool and go deep

Don't chase every new AI tool that launches. There are hundreds of them, and more every week. Pick one -- your coding agent, your writing assistant, whatever you use most -- and get really good at it.

The person who's excellent with one tool will outperform the person who's mediocre with ten. Depth beats breadth.

2. Build something every week

Even small things. A script that automates a tedious task. A landing page for an idea. A tool that solves a problem you have. The act of building is how you learn -- not reading about building.

This is the same advice I gave in Week 4, and it hasn't changed. If anything, it's more important now. The muscle memory of "describe, build, test, iterate" is the foundation for everything coming next.

3. Find a community

This is the one most people skip. Going alone is the slowest path. You learn faster when you can ask questions, share mistakes, see what others are building, and get feedback.

Not a Discord with 50,000 people where your question gets buried. A small community where people actually know each other and help each other. That's the environment where real learning happens.

---

Why I'm Building AI with Kian

I want to be honest about something. I started this blog series because I saw a gap.

The gap isn't between "knowing AI exists" and "not knowing." Most people know AI exists. They've seen the tweets, the TikToks, the LinkedIn posts from people who discovered ChatGPT and think they're visionaries.

The gap is between "knowing AI exists" and "actually using it to do something real." That's where most people get stuck. They read about AI, they watch videos about AI, they bookmark tutorials about AI. But they don't actually use it.

The eight weeks of this blog series were designed to close that gap. Each week: one concept, one action, one thing you can do by the end of the post. Not theory. Practice.

But here's what I've learned from the last two months: a blog isn't enough. The posts can teach you the concepts and walk you through the steps, but they can't help you when you're stuck at 11 PM with a weird error message. They can't review your code. They can't answer your specific question about your specific project.

That's where community comes in.

---

The Next Step

If you've been following along for the last eight weeks -- if you've set up the tools, built the projects, read the posts -- you've already done more than 95% of people who are "interested in AI." Seriously. Most people never make it past the bookmarking phase.

You're past that. You've shipped real things. You have real skills. The question now is: what's next?

If you want to keep going -- to go deeper into AI coding, build more complex projects, and have people to learn alongside -- the AI with Kian community is built for exactly that.

Here's what's inside:

• Weekly live sessions where we build things together in real time
• Code reviews -- share what you're building and get feedback
• A prompt library of real, tested prompts for common tasks
• Structured projects that build on what you learned in these eight weeks
• A small, active community of people who are actually building things (not just talking about it)

If that sounds like the environment you need to keep growing, you can join here.

No pressure. The blog posts will keep coming, free, every week. But if you want the live help, the feedback loop, and the people -- that's what the community is for.

---

The Eight-Week Recap

Before we go, look at what you accomplished:

| Week | What you learned | What you did |

|---|---|---|

| 1 | What AI coding agents are | Understood the landscape |

| 2 | How to set one up | Installed OpenCode + OpenRouter |

| 3 | Which models to pick | Benchmarked models for your needs |

| 4 | How to build something real | Built and deployed a landing page |

| 5 | The truth about prompt engineering | Focused on clarity, not magic words |

| 6 | How AI works in a real business | Saw real workflows with real numbers |

| 7 | The complete free tool stack | Set up $0/month AI infrastructure |

| 8 | Where this is all going | Prepared for what's next |

Eight weeks ago, you might not have known what an AI coding agent was. Now you've used one to build real software, deploy it to the internet, and understand enough to make smart decisions about tools, models, and workflows.

That's not nothing. That's a foundation. And the people who have this foundation are the ones who'll benefit most from everything coming next.

Stay building. Stay curious. And if you want company on the journey, the door is open.

---

What's the one thing you're most excited to learn or build next? I read every response.