Start.

I don’t start with apps anymore.

Not most of the time.

That is the simplest way I can describe how my work has changed with AI.

For years, knowledge work had a familiar shape. You opened the app where the work lived. The CRM for customer context. The dashboard for numbers. The doc for writing. The browser for research. Slack for the decision trail. Another tab for the thing you forgot.

Then AI arrived, and at first, most people used it at the edge of that workflow.

Write this email.

Summarise this note.

Clean up this paragraph.

Help me brainstorm.

Useful, but not the part changing my day.

The part changing my day is where the work starts.

I am starting in Codex

Right now, I spend most of my day in Codex.

That is not me trying to be dramatic about a tool. It is just what has happened.

When I need to write, review, research, build, organise, draft, inspect files, turn notes into something useful, or push a project forward, Codex is usually the place I start.

Not because Codex replaces every app I use.

It doesn’t.

I still need documents. I still need browser tabs. I still need internal tools. I still need dashboards, project systems, repositories, notes, and all the boring systems that keep work real.

But I do not always start there anymore.

I start with the task. Then I let Codex help me move through the systems around it.

That sounds subtle. It is not.

The old flow was app-first

The old flow looked like this:

Open the right app.

Find the right tab.

Search for the right context.

Copy something out.

Paste it somewhere else.

Ask AI for help.

Paste the output back.

Check what broke.

Repeat.

That was normal. We just accepted it.

The work was not hard because any single step was impossible. It was hard because the context was scattered everywhere. The thinking was in one place. The data was in another. The action happened somewhere else. The review happened in a fourth place.

AI as a chatbot helped, but it still sat outside the work.

You had to bring the work to it.

The thing I am noticing now is different. The AI is becoming a place where the work can begin.

The new flow is task-first

Now the flow feels closer to this:

Start with the task.

Bring in the context.

Let the agent read what matters.

Let it browse where needed.

Let it inspect files or tools.

Work through the draft, decision, fix or update together.

Keep human judgment where it belongs.

This is why tools like Codex, Claude, Cowork and ChatGPT matter so much.

They are not just smarter chat boxes. They are becoming places where work gets coordinated.

That matters because coordination is a huge part of modern work. Not the glamorous part. The real part.

Finding context.

Working across systems.

Knowing which app matters.

Remembering what happened last week.

Turning messy notes into a usable plan.

Moving from “I need to do this” to “here is the actual next step.”

That is where I think a lot of value is moving.

The apps still matter

I do not think this means the tools companies use disappear.

The CRM still matters.

The support platform still matters.

The analytics dashboard still matters.

The project tool still matters.

The CMS still matters.

The docs still matter.

Companies need systems of record. They need permissions. They need audit trails. They need shared history. They need places where work lands and can be trusted.

But I think the way people operate those systems changes.

More work starts in the AI tool, then moves through the existing tools from there.

A salesperson might start in Codex and ask for account prep. The agent reads the CRM, checks recent tickets, looks through notes, browses the company site, drafts a follow-up, and shows the human the decision points.

A support lead might start with a customer issue. The agent gathers the thread, checks the product context, compares similar issues, drafts the response, and points out what needs escalation.

A product manager might start with a rough idea. The agent reads past notes, checks open tickets, looks at usage data, drafts the product brief, and helps shape the tradeoffs.

The apps are still there.

But they are no longer isolated places a person clicks through alone.

They become systems a person and an agent move through together.

That is a product problem

This is where I think companies need to pay attention.

The question is not only “how do we add AI to our product?”

That is the obvious question. It is also the one everyone is already asking.

The better question is: what happens when the user brings an agent into your product?

Can the agent understand what is on the page?

Can it take action safely?

Can the human inspect what changed?

Can approvals happen cleanly?

Can the state of the app make sense to both the person and the agent?

Can the product handle a user who may not be clicking every button manually?

That is a different design problem from adding a chatbot.

It changes UI. It changes permissions. It changes logs. It changes rate limits. It changes error states. It changes support. It changes the shape of onboarding.

A product that works well for a human may not work well when a human brings an agent into the workflow.

That sounds early.

I do not think it stays early for long.

Agents still need people

This is the other part I think gets missed.

I am bullish on agents, but I am more bullish on capable people using agents well.

The useful future is not pressing a button and walking away from the work.

Someone still needs to know what good looks like.

Someone needs to give context. Someone needs to review the output. Someone needs to spot when the agent is confidently wrong. Someone needs to decide what should happen next.

That does not make the agent less valuable.

It makes the human more important in a different way.

The value shifts from doing every step by hand to directing the work, setting the standard, and knowing when to trust or challenge the system.

That is not “prompting” in the shallow sense.

It is operating.

It is knowing what outcome you want. It is knowing which context matters. It is knowing when the draft is good enough and when it is not. It is knowing when the tool should act, and when you need to slow it down.

That is a serious skill.

The timeline is closer than people think

I do not know whether this is 6 months, 12 months, 18 months or 24 months.

But I do think the direction is clear.

More people will start work inside tools like Codex, Claude, Cowork and ChatGPT.

Not all people. Not every task. Not every company at the same speed.

But enough that it changes what “using software” feels like.

The browser will still matter.

The apps will still matter.

The systems of record will still matter.

But the starting point changes.

And when the starting point changes, the shape of work changes with it.

What I would do now

If you are an individual, I would start experimenting with where work begins.

Pick one workflow this week. Something normal. Account research, meeting prep, project planning, content drafting, dashboard review, customer follow-up.

Do not start in the app.

Start in Codex, Claude, ChatGPT or whatever tool you use most.

Give it the task. Bring in the context. Let it help you move through the work. Watch what feels natural. Watch what feels broken.

That is how you build the instinct.

If you are a product person, I would ask a different set of questions.

What would an agent need to understand this page?

What actions should require approval?

What state changes need to be visible?

What would a useful agent-generated bug report look like?

What happens when one human user creates ten agent-driven actions in a minute?

These are not distant questions.

They are starting to become normal product questions.

I keep coming back to this because it is already changing how I work.

Codex is not a side tool for me anymore. It is close to the centre of my day.

That is why this feels less like a prediction and more like a direction of travel I am already inside.

The future of work will not arrive as a clean announcement.

It will start quietly, when more people stop opening the app first.

- JC