Meta is opening its AI models. What will set your company apart?

Welcome back.

Mark Zuckerberg published a 6,500-word manifesto yesterday describing Meta’s vision for AI. Here’s the gist.

He predicts that everyone will eventually have a personal agent that understands their goals, works continuously on their behalf, and helps them manage everything from their career to their health.

He argues that the greatest contribution of AI will be invention, not automation. People will be able to create businesses, products, scientific discoveries, and experiences that would have been too expensive or difficult to pursue before.

But the most consequential part of the manifesto is not Zuckerberg’s argument about who should control it.

Zuckerberg believes advanced AI should be distributed widely rather than concentrated inside a few companies or governments. Meta plans to offer free or affordable access to its agents, resume releasing open models, and make more of its technology available for other developers to build upon.

Most of the response will focus on whether this vision is optimistic, self-serving, or dangerously naive.

I think there is a more useful question for business leaders: Why is this particular vision of the future so attractive to Meta?

Meta’s real advantage

In The Future is for Everyone, Zuckerberg argues that concentrating advanced AI inside a few companies or governments would create an unhealthy imbalance of power. His proposed alternative is broad access.

He wants competing models and personal agents to check one another in much the same way that markets and democratic institutions distribute power. Meta will support this by releasing more open models and putting affordable personal agents into the hands of billions of people.

There is a genuine philosophical argument here. Broader access can increase competition, allow more people to examine how systems work, and give smaller companies capabilities that were previously available only to large institutions.

But there is also a very good business strategy.

Meta does not need to make most of its money by selling access to a model. It already owns some of the largest routes between technology and its users through WhatsApp, Instagram, Facebook, and its growing portfolio of AI glasses.

If models become cheaper, more open, and easier to replace, that strengthens Meta’s advantage.

Meta can let the intelligence layer become abundant because it owns much of the distribution layer above it.

That does not mean the manifesto is insincere. A principle can be sincerely held and strategically convenient at the same time. But it does reveal where Meta believes durable value will sit.

Not only in the model, but in the relationship around it.

The model is becoming a component

A lot of companies still describe their AI strategy through model and tool choices.

They are using ChatGPT, Claude, Gemini, or a particular open model. They have bought licenses, created an approved tool list, and started counting adoption.

Those decisions matter. Different models have different capabilities, prices, security arrangements, and limitations.

But access to a capable model is becoming easier for everyone. If your competitor can buy access to the same model tomorrow, the model itself cannot be the full source of your advantage.

The more interchangeable models become, the more value moves into what surrounds them: The context the AI can access, the workflow in which its output is used, and the trust people place in the result.

This is the distinction Meta’s strategy makes visible.

Meta wants intelligence to become widely available because it already has products, relationships, and distribution through which that intelligence can operate.

Most companies do not have Meta’s distribution. But they do have something Meta does not have. They have their own customer history, operating knowledge, decision rules, standards, and workflows. That is where their advantage can live.

The same model, two different systems

Imagine two customer success teams using the same AI model to prepare for renewals.

The first team gives employees access to the model. Each account manager uploads notes, asks for a summary, and uses the output however they see fit.

Some people get excellent results. Others barely use it. The quality depends on who is prompting, which files they remember to include, and how carefully they review the answer.

The second team builds AI into the renewal workflow.

The system assembles the current contract, account history, product usage, support issues, commitments, and prior renewal decisions. It flags conflicting information instead of resolving it silently. It cites the source behind every risk. The account owner reviews the brief before deciding what to do.

The teams may be using the same model. But they have not built the same capability. The first team has access to intelligence. The second has designed a dependable way of working around it.

If a better model appears next month, both teams can switch. But the second team keeps the most valuable part of what it built: the context, rules, integrations, review process, and learning from previous renewals.

The operating system around the model is not replaceable.

Four questions for your AI strategy

If you want to know whether your company is building something durable, ask four questions.

1. If we changed the model tomorrow, what would remain?

If the answer is nothing, you have adopted a tool. You have not yet built an organizational capability.

A useful AI system should retain value even when its underlying model changes.

2. What context does our system have that others cannot easily reproduce?

This might include customer history, internal standards, product knowledge, previous decisions, or feedback from completed work.

Raw information is not enough. It needs to be current, trusted, structured, and available at the right point in the workflow.

3. Where does the output enter a real decision?

An impressive answer sitting in a chat window has limited organizational value.

The output needs a destination, an owner, a review rule, and a consequence. It should help someone approve, prioritize, respond, create, or decide.

4. Does the workflow become better as we use it?

A durable system should generate evidence.

Which recommendations were accepted? Which risks were real? Where did the AI fail? Did the work become faster, cheaper, more accurate, or more valuable?

Without that feedback, the company may repeat the same mistakes with a more capable model.

Build above the model

Zuckerberg’s manifesto presents open AI as a way to distribute power.

Perhaps it will.

But open models alone will not make every company equally capable. They will make access to intelligence less scarce.

The harder work will remain.

Giving AI the right context.

Putting it inside a well-designed workflow.

Deciding where people must remain accountable.

Measuring whether the business is actually improving.

The companies that win will not necessarily be the ones that chose the perfect model earliest. They will be the ones that built systems capable of benefiting from whichever model is best next.

Start with one important workflow.

Identify which part is supplied by the model and which parts belong to your organization. Then strengthen the context, ownership, review, and measurement surrounding it.

Your model may change several times.

The capability you build around it should keep getting better.

Know someone making decisions about their company’s AI stack? Forward them this issue and ask what would remain if they changed models tomorrow.

We are working to make AI Ready more useful for the people responsible for these decisions. Share what you want us to cover in this short reader survey.

Haroon

Was this email forwarded to you? Subscribe to AI Ready to get the next issue.