Gemini 3.8 review: what it changes for AI web development

I've used Gemini 3.8 on real client projects since launch. Here's what it's great at, where it falls short, and how it compares to Codex and ChatGPT.

Gemini 3.8 changed AI web development video thumbnail by Ali Amin
On this page
  1. Watch the video
  2. What Google released
  3. Real project 1: a complex client CRM in one session
  4. Real project 2: 1,000+ pages from one prompt
  5. Working in Antigravity
  6. Gemini 3.8 vs Codex
  7. Where Gemini falls short: design
  8. The takeaway
  9. Follow along

Gemini 3.8 landed less than a month after Google's previous release, and it's built for agentic work: you give it a goal, and it works out the steps on its own. I've used it since launch on real client projects. Here's what Gemini 3.8 is good at, where it falls short, and how it compares to Codex and ChatGPT for AI web development.

Watch the video

Watch: Gemini 3.8 Just Changed AI Web Development

▶ Watch "Gemini 3.8 Just Changed AI Web Development (NEW UPDATE)" on YouTube

What Google released

Google shipped two models:

  • Gemini 3.8 Flash, available to everyone
  • Gemini 3.8 Flash Cyber, which you have to apply for on Google's website. It isn't open to the general public.

Gemini 3.8 leads on agentic benchmarks, and it's cheaper than the other models I use. Compared with Claude's Sonnet, the price difference isn't close.

Real project 1: a complex client CRM in one session

A client wanted to automate his email and see everything in one dashboard: inbound emails each day, pending reviews, quotations sent, revenue, replacements and more. I gave Gemini 3.8 one prompt plus the project files, and it built the whole structure in a single session.

With other models, I'd expect to spend more tokens and explain things in much more detail to get the same result. Gemini understood the overall picture without me pinpointing every detail. The next steps are connecting the database and the automations behind it. Here's how I approach custom CRMs as a GoHighLevel alternative.

Real project 2: 1,000+ pages from one prompt

I asked it for a website covering the US, and from a single prompt it produced 1,164 pages:

  • 50 state pages and 100 city hubs
  • 1,000 city service pages, each with its own copy
  • An SEO checklist applied to every page
  • Sitemaps, a clean header and footer, and a mega menu

Before building, it showed me a clear diagram of how the cities, states and services connect. Later I pushed the same idea further with ChatGPT Astra 6 and 10,000+ pages from one prompt.

Working in Antigravity

I use Gemini 3.8 inside Google's Antigravity. It shows you an implementation plan before it builds, and you can click anywhere on the plan to add a comment before you approve it. In all my time using it, I've never run out of tokens.

Gemini 3.8 vs Codex

I've used Codex for a long time too, but at the time of this video it had a bug that over-consumed tokens: a session limit that should last an hour was gone in 15 to 20 minutes. Even its speed mode didn't save me much time.

Gemini gives you code almost instantly and doesn't need every detail spelled out. For technical work right now, I wouldn't recommend Codex over Gemini.

Where Gemini falls short: design

Design is the weak spot. When I compared the same kind of site, ChatGPT's version had nicer logos, cleaner backgrounds and better images. ChatGPT has improved a lot at web design.

My current rule:

  • Design input: ChatGPT
  • Technical work: Claude or Gemini 3.8, not ChatGPT alone

If you're working in ChatGPT, here's which ChatGPT Astra 6 mode to use.

The takeaway

Gemini 3.8 is fast, cheap and strong at agentic, technical builds, especially big structured projects like CRMs and multi-page sites. Pair it with a model that's better at design, and you have a very capable AI web development setup.

If you want to learn how to use these models on real client work, join me inside AI Engineer Academy. Here's what's inside.


Follow along

Keep building.

Ali Amin

The community

Learn it with me, not alone.

AI Engineer Academy on Skool is where the courses, live calls and other builders are. Learn at your own pace, with people who are building too.