Here's the 80/20 of Ecom for this week:

In today's issue:

AI tools are a horse race.

If you asked me three months ago which AI tool was best for ecommerce sellers, I would have said OpenClaw.

Why?

Because of what you could do with it.

You could run autonomous AI agents, create sub-agents, schedule recurring workflows, analyze PPC data, monitor inventory, optimize listings, and start building an AI Chief of Staff for your ecommerce business.

Over the last couple of months, the competitors have caught up fast.

We’ve seen Claude put more pressure on third-party harnesses like OpenClaw.

We’ve seen Hermes surpass OpenClaw in a number of ways, including self-improvement, auto-creating skills, and better reliability.

And now we’ve seen the rise of Codex.

Right now, Codex has become my go-to AI super app.

I’m not saying Codex is the final answer.

This space changes too fast.

Maybe Claude upgrades next week. Maybe OpenClaw catches up. Maybe Hermes keeps improving.

But as of today, Codex is the 80/20 tool I would tell most ecommerce sellers to pay attention to.

It gives you a lot of what you can do with OpenClaw, but with much less setup.

For most sellers, that matters.

Because the problem is not usually:

The real question is:

There are three levels of AI use.

Level 1: I’m using ChatGPT like a Google search replacement.

Level 2: I’m using AI agentic apps like Claude Co-Work and Codex.

Level 3: I’m using OpenClaw and/or Hermes and have my own AI Chief of Staff.

The difference between Level 1 and Level 2 is this:

At Level 1, you are still copying and pasting back and forth.

At Level 2, AI can actually do work for you.

That is the shift.

Not just:

But:

Or:

Or:

That is where this is headed.

Do me a quick favor:

Hit reply and let me know which level you are at:

1, 2, or 3.

I’m paying attention to Codex for a few reasons.

OpenClaw is powerful.

I still like it.

But setting up a VPS, dealing with dependencies, keeping things running, debugging errors, and making sure agents don’t break is not easy.

If you are technical, that is manageable.

If you are an ecommerce operator juggling inventory, PPC, sourcing, Amazon support, cash flow, and team management, it can be too much.

Codex gives you maybe 60–80% of what most sellers want from OpenClaw, but with around 20% of the setup.

That is why I’m calling it the 80/20 tool right now.

Most people are not using OpenClaw to its fullest extent anyway.

If Codex gets you moving faster, it may be the better first step.

Claude is powerful, but the experience is split into different parts:

Each one is useful.

Moving between them can feel clunky.

You can’t always move work easily from one place to another.

Codex feels more unified.

You can chat, launch agents, create skills, run workflows, use a browser preview, generate images, and even control your computer from inside one app.

That matters because ecommerce work is messy.

You are looking at CSV files, Seller Central exports, product images, competitor listings, email drafts, SOPs, and reports.

A tool that can move across those pieces is more useful than one that handles a narrow slice.

A few months ago, if you asked me which model was smartest, I would have said Opus.

Hands down.

It was ahead of the others.

Right now, GPT 5.5 has caught up enough that I’m comfortable using it daily.

I still use Opus for certain things.

Claude often explains things better.

It can feel more articulate.

But be careful.

There is a real thing called fluency bias.

Just because AI explains something smoothly does not mean it is right.

One of my OCEA students said this well on today’s call:

That is a good rule.

For important work, I still like using more than one model:

AI is the analyst.

You are still the operator.

This is one of the most important ideas.

A skill is basically an SOP for an AI agent.

It tells the AI:

Think of Neo in The Matrix.

They plug something into the back of his head, and suddenly he knows kung fu.

That is what a skill does for an AI agent.

Instead of explaining your PPC workflow from scratch every time, you package it as a skill.

Then the AI can reuse it.

For example, I took one of the OpenClaw PPC skills we built and asked Codex to adapt it.

It did.

Then I tested inventory management.

That worked too.

That means many of the ecommerce workflows we’ve been building in OpenClaw can be moved into Codex.

Not always perfectly.

You still need to verify.

But the direction is clear.

Workflows are becoming more portable.

Nano-banana fans beware. There's a new king in town. As of a few weeks ago, GPT-5.5 has the best-in-class AI image generator in ImageGen2. Don't take my word for it. I spoke recently with my friend Kamal Singh, who leads Listing Optimizer.AI, which is the de facto standard in creating AI-optimized Amazon images. He agrees that NanoBanana is no longer the best AI generating model. It's actually ImageGen2, which is already built in Codex.

Check out the images I generated in a PickFu skill that I used entirely in Codex to generate images. Run PickFu polls against competitors and generate multiple rounds of Amazon main images to eventually beat the competition. All of this can be done entirely within Codex. Check out my YouTube video walkthrough here.

Here are three practical workflows ecommerce sellers should watch.

Most sellers are sitting on PPC waste.

The problem is not lack of data.

It is that nobody has time to review it properly every week.

Codex can review search term reports, campaign exports, bulk files, and performance data to flag where money may be leaking.

The value is not just “AI tells you to lower a bid.”

The value is having an analyst look across the mess and surface the few places that need your attention.

For example:

I would not let AI execute this blindly.

But as a first-pass analyst, it is very useful.

Inventory mistakes are expensive.

Stock out, and you lose ranking.

Over-order, and you tie up cash.

Move too slowly, and you miss the window.

This is where an AI agent can be an early warning system.

It can look at sales velocity, inventory levels, lead times, promotions, and SKU-level risk.

The goal is simple:

Show the owner what needs attention before it becomes a cash flow or stockout problem.

That alone can be worth a lot.

Most listing optimization is too shallow.

People ask AI to rewrite bullets.

That is not the real opportunity.

The real opportunity is a full listing diagnosis.

Codex can look at:

Your listing does not win or lose because of one bullet point.

It wins or loses because of the whole system.

Images, copy, keywords, price, reviews, and competition all affect conversion.

A good AI workflow can show you what is holding the listing back.

The most interesting Codex feature I’m testing is the /goal command.

You know how sometimes AI feels like an employee who keeps stopping?

You ask it to do something.

It gives you one answer.

You say, “Okay, now go do it.”

It gives you another small step.

Then you have to keep prompting it.

That gets old fast.

/goal changes the pattern.

You give the agent a finish line.

Then it keeps working until it reaches the goal, gets stuck, or cannot continue.

That matters because a lot of ecommerce work is not one question.

It is a sequence.

A listing analysis is not one question.

A PPC audit is not one question.

A product opportunity review is not one question.

An inventory risk report is not one question.

These are workflows.

That is where AI is moving.

The future of AI in ecommerce is not better prompts.

It is better workflows.

Do not blindly trust the AI.

One of the biggest mistakes ecommerce sellers will make is assuming that because AI sounds confident, it must be right.

That is dangerous.

Especially with:

Use AI to analyze.

Use AI to draft.

Use AI to compare options.

Use AI to find problems faster.

Keep a human in the loop for anything that can affect revenue, rankings, cash flow, account health, or customer trust.

AI should make you a sharper operator.

It should not replace your judgment.

Next week, keep an eye out for my email where I'll show you how to implement this into your own Amazon business. Stay tuned

If you are doing at least $500,000 a year in ecommerce, here is what I would pay attention to:

Find one workflow in your business that already hurts.

Not ten workflows.

One.

For most sellers, it is one of three:

That is where the ROI is.

Then define the workflow simply:

The sellers who win with AI will not be the ones chasing every new tool.

They will know which workflows matter.

Ritu Java wrote in her latest AI for E-Commerce Newsletter about AI video and Amazon Ads workflows.

The part that stood out was not just the AI creative angle.

It was the guardrail.

If AI can connect to your live ad account, that does not mean it should make changes directly.

For PPC, I like this sequence:

That is boring.

That is how you avoid waking up to expensive mistakes.

TAKEAWAY: For PPC, the file is the guardrail. AI can help you move faster, but live account changes need human review.

PPC Ninja:

https://www.ppc-ninja.com/

Marketplace Pulse had a useful macro point this week: ecommerce growth can look strong on paper while feeling weak for operators.

Revenue growth can hide a lot.

Higher prices, paid traffic quality, BNPL usage, repeat buyer behavior, cash flow, and inventory pressure all matter.

This is another reason I like weekly AI-assisted CEO reports.

Not 50 charts.

Just the signal.

TAKEAWAY: Do not only ask, “Did revenue go up?” Ask what changed underneath it.

Talk soon,

Gary