Astra, Fable and the New AI Models Are Changing What Counts as AI Work

Every time a new AI model launches, the internet immediately divides into three groups:

1. People declaring that everything has changed.
2. People insisting that nothing has changed.
3. People frantically refreshing their accounts because everyone else seems to have access and they do not.

GPT-6 Astra arrived last week, alongside Claude Fable 5.1, and I will confess that my initial reaction fell squarely into Group One.

Because Astra is not simply a smarter version of ChatGPT.

It represents a much more important shift: AI is moving from generating pieces of work to completing workflows.

That distinction may sound subtle. It is not.

We Are Moving Beyond “Help Me Create This”

Most people still use AI one task at a time:

“Write this email.”

“Summarize this report.”

“Give me ten campaign ideas.”

“Create an outline for this presentation.”

The problem with that approach is the human is still responsible for moving the work through every stage, finding the information, feeding it to the model, checking the output, transferring it into the right format and deciding what happens next.

Astra is designed to operate across multiple steps, tools and applications. It can research, analyze information, use software, create finished documents and check its own work while keeping the larger objective in view.

In other words, the prompt is no longer necessarily:

Write the presentation.

It can become:

Review these research reports, identify the five findings our executive team needs to understand, compare them with our current strategy, build the presentation using our corporate template, check every statistic against the source material and flag the decisions leadership needs to make.

That is not content generation.

That is a work assignment.

And that is why Astra feels different.

What This Could Look Like in Marketing

Imagine a marketing team preparing for a quarterly business review.

Today, someone may spend hours gathering campaign results from different files, looking through customer research, copying numbers into a spreadsheet, identifying what changed and turning it all into a presentation.

With Astra, the team could provide the relevant sources, the reporting requirements and the company’s presentation template. The AI could then:

• Review the campaign data.
• Compare performance across audiences and channels.
• Identify meaningful changes rather than simply repeating the numbers.
• Investigate possible explanations.
• Build charts and tables.
• Create an editable executive presentation.
• Check the completed deck against the original sources and brand template.
• Flag the questions that still require human judgment.

OpenAI specifically positions Astra for demanding, end-to-end work involving sustained reasoning and multiple tools. Its ability to carry context across the workflow, and incorporate new direction without forgetting the original goal, is one of its most consequential improvements.

That does not eliminate the marketing team. It changes where the team spends its time.

Instead of spending hours assembling the evidence, the humans can interrogate it:

• Are we interpreting this correctly?
• What are we overlooking?
• Does this recommendation fit what we know about our customers?
• What decision should we make?

That is a much better use of human intelligence than manually transferring numbers from Column G into Slide 14.

For years, in nearly every AI training I’ve led, I’ve emphasized exactly that idea. People should use AI to do the things they hate doing – which is nearly always busy work that requires no creativity. Things like compiling data and yes, transferring numbers from a spreadsheet into a Powerpoint. GPT-6 Astra helps them do exactly that.

It Can Also Produce the Actual Deliverable

There is another important change here.

Until recently, AI could provide surprisingly strong content while still producing deeply mediocre deliverables. You’ve seen it yourself, I’m sure. It might write the words for a presentation, but a human still had to build the presentation. (Have you ever tried to work with your brand’s templated deck while using AI? It’s a nightmare!)

The latest OpenAI models are designed to create and refine editable presentations, documents and spreadsheets, not simply describe what those files should contain. Astra adds stronger reasoning and judgment to that production capability.

Now, to be clear, even GPT-6 Astra struggles to understand branded template structures. But, it’s better than previous versions.

In one demonstration published by OpenAI, Thomas Ricouard used Astra in Codex to translate a simple house brief into an editable 3D scene in Blender. He then worked with the model to refine the architecture, furnishings, materials and lighting before developing an Unreal Engine walkthrough.

No, most marketing teams are not going to spend next Tuesday building virtual houses.

But consider what the example actually demonstrates: Astra can translate an idea into a complex, editable artifact, inspect the result and continue refining it through multiple stages.

Now imagine giving Astra a communications workflow, supporting documents, brand standards and review requirements, and asking it to turn that information into an interactive Human + AI Workflow Blueprint. The result would not simply be a written recommendation. Astra could build the working tool, allow the team to test it and then refine the experience based on their feedback.

The same capability could be applied to:

• An interactive product experience.
• A customer-facing ROI calculator.
• A visual market-intelligence dashboard.
• A fully formatted campaign planning workbook.
• A stakeholder portal following a merger.
• A working prototype of a new digital service.

A marketer does not need to know how to build every underlying component. But the marketer does need to understand the audience, define the objective and recognize what a successful result looks like.

That last part is becoming increasingly important.

You Still Need to Choose the Right Model for the Work

Astra is the most capable OpenAI model, but that does not make it the right model for every assignment.

Using the most powerful model to reformat 300 product descriptions would be like hiring a management consulting firm to alphabetize the supply closet. I suppose they could do it. The resulting PowerPoint explaining their methodology would be lovely.

OpenAI’s current model family is essentially divided by the complexity of the work:

• Astra is intended for the hardest end-to-end assignments involving multiple steps, tools and decisions.
• Sol is designed for complex, open-ended work requiring significant analysis and polish.
• Terra is the practical model for everyday professional work.
• Luna is suited to clear, repeatable, high-volume tasks such as extraction, classification and structured summaries.

That means the question is not, “Which model should our company use?”

The better questions are:

• What work are we trying to accomplish?
• How much reasoning does it require?
• What information and tools must the AI access?
• What could go wrong?
• Where must a human make the decision?

Organizations do not need one AI model. They need a thoughtful model strategy.

And No, Claude Did Not Suddenly Become Irrelevant

Claude Fable 5.1 also arrived with significant improvements in long-running knowledge work, research, coding and computer use. Anthropic says the model is better at avoiding shortcuts, finding root causes and maintaining coherence during complex assignments that may run for hours.

In testimonials published by Anthropic, a senior portfolio manager at Millennium said Fable 5.1 traced an extremely rare, years-old software crash to a bug in an external vendor library. Plaid reported that it mapped a change spanning eight services and three codebases, down to individual functions and database records. Canva also highlighted improvements in writing quality and the model’s ability to follow specific writing guidance.

That matters.

I would not recommend that an organization switch its entire AI environment every time one company wins a benchmark or LinkedIn collectively develops a new favorite model.

That is not an AI strategy. It is model FOMO with a procurement department.

Run your own evaluations using your organization’s actual work.

A communications team may discover that it prefers Fable’s writing for certain executive materials while Astra performs better on a research-to-presentation workflow. A high-volume content operation may get everything it needs from Terra or Luna. A company deeply embedded in Google Workspace may find Gemini’s native ecosystem advantages more valuable than a small difference in model performance.

Gemini 3.5 Flash is also pushing further into agentic work, including multimodal document processing, computer use and enterprise workflows that continue across multiple steps. Google has shown it being used for complex invoice processing, ongoing administrative workflows and agents that work across business systems.

There is no universal champion because there is no universal workflow.

The Real Competitive Advantage Is Not Access to Astra

Your competitors can buy access to the same models.

The advantage comes from knowing how to redesign work around them.

That requires much more than teaching employees a few impressive prompts.

Teams need to determine:

• Which workflows are valuable enough to redesign.
• Which parts AI can support or complete.
• Which sources the AI should be allowed to use.
• Which decisions must remain with humans.
• How completed work will be reviewed.
• What evidence the AI must provide.
• How knowledge, prompts and successful processes will be shared.
• Which model provides the right balance of capability, speed, risk and cost.

Astra makes much more ambitious work possible.

It also makes weak operating practices more dangerous.

If you give an AI responsibility for an entire workflow without defining the standards, evidence, approvals and stopping points, you have not created an intelligent system. You have created a very fast way to scale ambiguity.

This Is the Shift Leaders Need to Understand

The most important development is not that Astra can write a better strategy or build a prettier presentation.

It is that the boundary between using AI and delegating work to AI is disappearing.

We are moving from: “Help me perform this task.”

To: “Take this assignment, use these resources, follow these rules, create the deliverable, check your work and return to me when a decision requires human judgment.”

That changes job design.

It changes workflow design.

It changes training.

It changes governance.

And it changes what leaders should be asking of their teams.

The question is no longer simply, “Are our employees using AI?”

The question is: Have we designed how humans and AI should work together now that AI can do more than wait for the next prompt?

Astra may be the strongest example of that shift today.

But if the last few weeks have taught us anything, “today” is doing an enormous amount of work in that sentence.

---

Remember: Buying technology is easy. Changing how people work is harder.

Human Driven AI helps organizations move from scattered AI experimentation to shared, scalable practice. We build the governance foundations, redesign workflows around human and AI strengths, and deliver custom training programs that teach your teams how to put those systems into practice.

Whether you need an executive Lunch-and-Learn, a hands-on offsite workshop, an enterprise AI transformation program or a strategy to strengthen your brand’s visibility through GEO, we help you turn AI capability into better ways of working.

Ready to move from using AI to working differently with it? Let’s talk.

Read more: Astra, Fable and the New AI Models Are Changing What Counts as AI Work