Astra Is the AI Shift Agencies Should Actually Care About

From assistance to delegation

I’ve spent the last three days using ChatGPT Astra almost nonstop, and after three years of using AI, this is the first release that has made me seriously rethink the operating model of an agency.

Astra can do the production work, not just help you do it.

That is a much bigger change than another model getting 20% better at writing, coding, research, or design.

For the last three years, AI has mostly been a force multiplier for someone already doing the work. A copywriter uses ChatGPT to write faster. A developer uses Claude to code faster. A designer uses an image model to generate an asset faster.

With Astra, you give it an outcome, it operates the computer, moves between websites and software, completes the steps, and comes back when it actually needs a decision from you.

That has HUGE implications for agencies because most agency employees spend far more time producing work than deciding what work should be produced.

That dynamic needs to radically change.

The 3 shifts agency owners need to understand

I’ve been obsessed with AI since ChatGPT launched because it was obvious where the technology was heading.

Agencies sell knowledge work. Strategy, research, writing, design, development, analysis, operations, project management, and account management are all knowledge work in one form or another.

AI has already changed pieces of those jobs.

Astra can start taking over the workflow around those pieces too.

I think there are three changes that matter most:

1. AI is moving from a tool an employee uses to a worker an employee manages.

2. The biggest productivity gain will happen at the frontline, not the executive level.

3. As production becomes easier to delegate, strategy, creativity, communication, taste, and judgment become much more valuable.

Put those together and the job description of an agency employee starts to look very different.

Astra turns AI into someone you can delegate to

Here is a very simple example.

I wanted to update the VantaFive website and LinkedIn presence. Historically, that would mean either doing the work myself or handing the project to someone else.

Someone would need to review the existing website, understand the brand, create the assets, find the portfolio company logos, update the site, upload the LinkedIn banners, and check everything when they were finished.

AI could already help with individual steps.

ChatGPT could help with the copy.

An image model could help create an asset.

Claude could help with code.

But a human still had to sit in the middle and move the project from one step to the next.

With Astra, I basically gave the same instructions I would give an employee.

I told Astra what I wanted changed, gave it the relevant context, answered a couple of questions, and let it go do the work.

Thirty minutes later, the changes were live.

I did not get a banner file that I still needed to download and upload myself. Astra created the banner, accessed the page, uploaded the banner, and finished the job.

Most agency work is not one task. Agency work is a series of tasks connected together.

Creating an ad is not just writing the copy. Someone researches the account, writes the copy, builds the creative, uploads the campaign, checks the settings, launches the campaign, and monitors the result.

Creating a client report is not just analyzing the data. Someone pulls the data, cleans the data, formats the data, puts the data into slides, writes the explanation, shares the report, and updates whatever internal system tracks the account.

The production work surrounding the actual thinking adds up quickly.

Astra can increasingly own that production layer.

This changes the skill required to use AI

For the last few years, everyone has talked about prompt engineering.

I think prompting becomes much less interesting as these systems get better.

I am not particularly technical. I don’t build MCP servers. I haven’t spent years learning complicated AI workflows. I can explain an outcome to Astra the same way I would explain the outcome to a smart employee, and Astra can figure out a surprising amount of the rest.

The skill that matters is delegation.

Can you clearly explain the outcome?

Can you provide the right context?

Can you define the constraints?

Can you decide where Astra needs approval and where Astra can make the call?

Can you review the finished work and know whether the work is actually good?

Those are management skills, not AI skills.

Frontline employees get the biggest productivity gain

Astra is useful to me, but I don’t think CEOs are going to get the most leverage from it.

CEOs already have people to delegate to.

If you run a reasonably sized company, you probably have directors, assistants, accountants, lawyers, operators, salespeople, and functional leaders around you.

A large percentage of your job is already deciding what needs to happen and getting other people to make it happen.

The frontline employee has a very different job.

There is nobody below the frontline employee on the org chart.

The strategist comes up with the strategy and then builds the deck.

The account manager figures out what the client needs and then updates the CRM, project management system, notes, reporting, and follow-up.

The marketer determines what campaign should run and then builds the list, creates the assets, uploads the campaign, QA’s the campaign, launches the campaign, and pulls the results.

Frontline employees are where the work actually stops and gets produced.

Astra effectively gives every one of those employees someone to delegate production work to.

Take an agency role that is 20% strategy and 80% production

If Astra can eventually handle most of the 80%, the employee does not become 20% more productive.

You are changing what that person spends most of the week doing.

Someone who previously spent one day thinking and four days executing can now spend most of the week thinking.

If a role is closer to 50% strategy and 50% production, the same logic applies. Remove a large portion of the production layer and the amount of time available for high-value thinking can potentially double.

I want to be very clear about what I am not saying.

I am not saying every agency should look at a person who can now do five times as much work and immediately give them five times as many clients.

That would be the dumbest possible implementation of this technology.

The opportunity is to change what the person spends their time doing.

If your best strategist currently spends 60% of the week building decks, formatting documents, pulling research, and updating systems, Astra should increasingly own that 60%.

Your strategist should spend that time talking to clients, understanding the problem, developing better ideas, improving the strategy, and reviewing the work Astra produces.

Production gets cheaper. Judgment gets more valuable.

I still think AI is bad at a lot of things.

AI is still bad at taste and nuance. It does not understand the weird interpersonal context behind why a client is actually upset. It often misses when the technically correct answer is still the wrong answer. And I don’t trust it to decide which of five reasonable strategies makes the most sense for a specific company at a specific moment.

I will die on that hill.

What AI is becoming incredibly good at is production and execution.

Agencies currently bundle those two types of work together.

We hire a great strategist, then make the strategist spend hours formatting decks.

We hire a great marketer, then make the marketer spend hours uploading campaigns and moving information between software.

We hire an experienced operator, then make the operator spend half the week maintaining processes that could increasingly be maintained automatically.

That model made sense when production required human labor.

It makes less sense when production can be delegated.

Your best people become MORE valuable

People hear “AI can do the production” and immediately jump to “so we need fewer talented people.”

In some cases, yes. But I think the bigger effect is that your best people become much more leveraged.

A top 1% strategist who can only oversee five major client strategies at once has one level of value.

A top 1% strategist who can oversee 15 because Astra handles the research, formatting, documentation, system updates, and other production work is worth something completely different.

This is why I have been saying that the most valuable person inside an agency is not someone who is “good at AI.”

It is someone who is world-class at strategy and knows how to use AI as leverage.

You still need the person to be exceptional at the actual job.

Every employee is now a manager

Historically, individual contributors learned how to execute.

Then they got promoted and learned how to manage.

That sequence is about to get weird.

A 24-year-old individual contributor using Astra now needs to know how to define an outcome, break a project into pieces, communicate context, delegate execution, establish guardrails, review work, give feedback, and know when to intervene.

Those are the same skills we expect from a good manager.

So every employee effectively becomes a manager.

Not necessarily a manager of people.

A manager of work.

And I think agencies are going to need to completely rethink training around that.

The employee who is great at manually completing a process but terrible at explaining the process is going to struggle.

The employee who understands why the process exists, what a good outcome looks like, what can go wrong, and where human judgment is required will thrive.

Instead of asking:

“Can you do this?”

The question becomes:

“Can you manage this to the right outcome?”

That is a much higher bar.

What I would do if I ran an agency right now

I am still figuring this out inside our own companies, so I don’t have some perfect AI operating model to hand you.

But I would start doing five things immediately.

1. Break every role into strategy vs. production

Take every major job inside your agency and actually look at how the person spends the week.

Separate the work that requires strategy, creativity, communication, relationships, taste, or judgment from the work that is research, formatting, data entry, system maintenance, scheduling, asset production, reporting, and repetitive execution.

Then put a percentage against each side.

If a role is 70% production, I would be spending a lot of time figuring out why a human still owns all 70%.

2. Redesign the job instead of adding an AI tool

“Everyone gets ChatGPT” is not an AI strategy.

If you give an employee Astra and the employee still has the exact same job description, exact same workflow, exact same utilization target, and exact same responsibilities six months later, you probably missed the point.

Don’t just use Astra to make the existing job faster. Figure out what the job should become once Astra handles a significant amount of production.

3. Teach individual contributors how to delegate

Your frontline people need to get good at the same skills you traditionally teach new managers.

They need to define clear outcomes, provide context, establish constraints, decide where approvals are required, review work, and give useful feedback.

I would spend less time teaching people the perfect prompt and more time teaching people how to manage work.

4. Move senior people back toward thinking

Senior agency talent is expensive.

If your $200K strategist is spending 15 hours a week building slides, gathering screenshots, moving information between tools, and formatting deliverables, your problem is not that the strategist needs to work faster.

Your problem is that you are paying senior talent to do production work.

Astra gives agencies a chance to unwind a lot of that.

Keep your best people on the work that requires their brain.

5. Put real guardrails around security

There is an obvious downside to software that can operate your computer.

Astra can access whatever you give Astra access to.

I am personally being careful around financial accounts, confidential information, and sensitive company data. Agencies using computer-operating AI at scale will need clear permissions, approval processes, access controls, and rules around what the AI can touch.

But I also don’t think “there are risks” is a viable reason to sit on the sidelines while your competitors figure out how to use the technology safely.

Why I think agencies need to move fast

I normally hate the “adapt or die” language people use every time a new technology launches.

Most technology adoption takes longer than Twitter makes you think.

I am less confident that agencies have that luxury here.

Agency economics are largely labor economics.

A huge percentage of an agency’s costs come from the people required to produce the work.

Now imagine two agencies selling essentially the same service.

The first agency uses AI to occasionally make existing employees faster.

The second agency rebuilds the entire delivery model around the assumption that production work can increasingly be delegated to Astra.

That second agency can put more of its payroll toward people with great strategy and judgment. It needs fewer hours to produce the same output, which creates room to increase gross margins, lower pricing, invest more in growth, or simply deliver a much better product at the same price.

Its best people also spend more time thinking and less time producing.

Run those two businesses next to each other for 12 months and I don’t think they look remotely the same.

For agencies, that creates an unusually big gap between the firms that redesign delivery around AI and the ones that just give everyone another tool.

If your competitor figures out the operating model first, the problem is not that their employees have a slightly better AI tool.

The problem is that they have built a fundamentally more efficient agency.

The question I think every agency owner should be asking

For the last three years, everyone has asked some version of:

“How can AI make my employees faster?”

I think Astra creates a much more important question:

“What should my employees actually spend their time doing now that they can delegate most of the production work?”

The answer affects who you hire, how you train them, which employees become most valuable, how you structure delivery, and eventually how the agency makes money.

I’ve believed since ChatGPT launched that AI would eventually force agencies to rethink how knowledge work gets done.

After using Astra for the last three days, I think we are finally getting to that point.

Every frontline employee just got someone to delegate to.

Agency owners now have to decide what they are going to do with all of that leverage.