#8: Image Generation

This is a huge shift.

For months, Nano Banana has been the king of AI image generation.

Just last week, that’s finally changed.

ChatGPT just launched Images 2, and it’s shockingly good.

Real use cases: marketing campaigns, photoshopping, realism, crafting entire design systems, panoramic worlds.

You can use Images 2 inside ChatGPT now - just make sure to include “use ChatGPT Images 2 to make this…” in your prompts!

#7: Video Generation

Most of the AI-generated videos on short-form platforms like TikTok are coming from Seedance2.0.

And that’s because the 2.0 update to Seedance is extremely good at executing realism, camera cuts, lighting, and facial expressions.

Your simplistic media stack should be:

Images 2 for image generation, and then upload those images to Seedance2.0 to generate full-stack videos.

If you haven’t tried it yet, you can here: https://seedance2.ai/

#6: Coding

It’s no surprise that we make a LOT of Claude content - especially Claude Coding-related content.

But with the release of GPT-5.5, we have to admit that for the first time in a while, GPT-5.5 inside Codex is simply superior for 99% of coding tasks.

We’re still using Claude Code, and the entire Claude experience is great - but Codex really shines for browser tasks, simplicity, and we especially recommend it if you’re already familiar with GPT tools.

Try here: https://developers.openai.com/codex/app

#5: AI Memory System

One of the biggest problems with AI is that it frequently forgets context and important data that’s needed to get the LLM responses that you want.

You can fix this issue by building a permanent AI second brain with Claude Code and Obsidian.

Your second brain stores all your projects, preferences, AI instructions, and more.

You then connect your second brain to any AI tool and get instant memory syncing across AI tools.

A must-use for anyone who relies on multiple AI tools or simply wants better memory.

We teach you how to build an Obsidian second brain with Claude Code here:

#4: AI-Powered Design

Claude Design is the most powerful AI-powered design tool ever.

You can build entire websites, marketing campaigns, and any other visuals you may need in seconds.

We recommend you build “Design Systems” which are like Claude Skills, but for visuals.

They load all your color, font, and design preferences so Claude can build perfect visuals.

If you haven’t tried Claude Design yet, just go to Claude → Left Sidebar → Design.

#3: Cost Efficiency

DeepSeek V4 is live now, and it’s currently the best LLM for cost efficiency.

On average, it’s ~70% cheaper than the top LLMs from OpenAI, Anthropic, and Google.

Let’s be honest: Don’t expect Opus 4.7 or GPT-5.5-level intelligence from DeepSeek models, and in general, DeepSeek models can feel robotic.

But, that’s not the role DeepSeek should play in your AI stack.

We recommend DeepSeek for: Spare OpenClaw setups, agentic workflows, easy coding tasks, math, and overall, a “different” experience than your typical LLM from OpenAI or Anthropic.

V4 is extremely fun - try it here: https://chat.deepseek.com

#2: Agentic Workflows

For 80% of automated tasks, Manus is our first recommendation.

It can connect to all your tools (Gmail, Google Workspace, etc.) and is very easy to use out of the box.

Manus is extremely good at Email management, general research, deep-dive analysis, creating presentations, and financial research.

A must-use tool for anyone trying to level up with AI: https://manus.im/app

To round off our list, we had to include Opus 4.7 as the best general AI on the market right now.

It’s excellent at coding, its personality is “warm", it connects to all your tools, has access to Claude Skills, and much more.

With the recent release of GPT-5.5, many of our coding tasks have been migrated, but for general use, we’re relying more than ever on Opus inside Claude.

As discussed above, GPT-5.5 is now live!

It’s the best model in the world, and for the first time in months, we’re switching key workflows back to ChatGPT from Claude.

Watch Miles’ Ultimate GPT-5.5 guide here:

Most people are completely overlooking DeepSeek right now.

For a fraction of the cost, you’re getting GPT-5.4/Opus 4.6 performance, which is still extremely powerful.

For basic tasks where you don’t want to use your OpenAI/Anthropic account tokens, we recommend relying on the new DeepSeek V4.

Full benchmarks & comparison:

Persistent AI agents that operate within a team's ChatGPT workspace - handling ongoing tasks, without needing to be re-prompted each session.

Our advice: Workspace agents are only available to enterprise plans - if you want to try them, you can purchase a business account for $20/month, which actually might be a better value than a standard GPT subscription.

Official announcement:

https://openai.com/index/introducing-workspace-agents-in-chatgpt/

For those who like some good drama.

The Musk vs OpenAI trial began yesterday (April 28).

Elon Musk claims Sam Altman and OpenAI enriched themselves, made themselves more powerful, and breached the very basic principles on which the original OpenAI non-profit was founded.

The most powerful GPT-5.5 Vibe Coding prompt on the entire internet.

Use this prompt, and you'll literally be able to ship anything.

Fully functional apps, beautiful web designs - whatever you can imagine, this prompt can execute.

The best part is, it takes <5 minutes to use:

Y-Combinator just published its request for AI startups for Summer 2026.

For those of you looking to make money with AI and start an AI-powered business, there are some good startup ideas in here:

(summary from their site)