How AI agents could change the way companies are organized

Artificial intelligence is on its way from being a tool employees use to becoming something that can carry out tasks, work with people, and coordinate entire parts of a company. That could have far greater consequences than simply working faster. It could change who does the work, who leads it, and how companies are organized.

In the article The Future of the Firm, OpenAI employees Hemanth Asirvatham and Elliott Mokski describe how artificial intelligence could change the way the companies of the future are organized. They present three possible developments that challenge the traditional way we think about employees, managers, and departments.

This is not just about AI being able to write emails, analyze spreadsheets, or automate customer service. It is about what happens when AI agents can take over larger areas of responsibility, coordinate tasks, and work with both people and other AI agents.

The authors describe possible developments, not a final answer to what the companies of the future will look like. They also stress that these are their own assessments, not OpenAI's official position.

Beneath us. When every employee gets their own AI coworkers

The first development is perhaps the easiest to imagine. The company keeps the structure it has today, but every employee gets access to AI agents that do more and more of the work.

Imagine a buyer at an IT company. Today, this person spends a lot of time gathering prices, comparing suppliers, checking inventory levels, following up on orders, and looking into delivery problems. These are tasks that often require several systems, emails, phone calls, and manual checks.

With AI agents, much of this can be done automatically. One agent gathers quotes from suppliers, another compares prices and delivery times, while a third follows up on orders and spots possible delays. The agents can work together and give the buyer a combined overview to base decisions on.

The buyer no longer needs to spend as much time gathering information. Instead, they can focus on supplier strategy, negotiations, profitability, and decisions that require experience and judgment.

The difference is that the buyer goes from doing all the tasks themselves to managing the work the AI agents do.

The same could happen in marketing. A marketer who used to spend several days researching the market, planning campaigns, creating content, and analyzing results could have AI agents do much of this. The marketer sets the direction, assesses the quality, and makes the decisions.

This could mean that employees become responsible for far more than before, without the company necessarily needing more people.

The authors describe it as every employee, in a sense, getting promoted. Not because job titles change, but because they get digital coworkers under them and can therefore take responsibility for larger tasks.

Does this mean the company needs fewer employees?

Not necessarily. That is an important point in the article.

If a company becomes far more productive, it can choose to produce more, develop more products, serve more customers, or enter new markets. Then employees could still have just as much to do, but with different and larger areas of responsibility.

A marketing department that used to manage four campaigns a month might be able to run twenty. A sales department could follow up with far more customers, while a purchasing department could handle a larger number of suppliers and products.

But there is also another possible outcome. If the company has no need to produce more, or the market doesn't demand more, higher productivity could mean it needs fewer employees.

So AI does not guarantee that jobs will be preserved. The deciding factors will include how much more the company wants to achieve, what customers demand, and what value people still add.

The authors also point to a challenge for recent graduates. Many have traditionally started with simple, routine tasks to learn how a company works. When AI takes over much of this, companies will have to find new ways to give young employees experience and practical understanding.

Beside us. When AI agents become coworkers on equal footing with people

In the second scenario, the development goes one step further. Now each employee doesn't just have their own AI agents. AI also becomes part of the organization itself, with its own areas of responsibility in the same way as human employees and departments.

Imagine a company that is about to launch a new product.

Usually this requires cooperation between marketing, sales, purchasing, and logistics. The marketing department has to plan the campaign, the sales department has to prepare for customer contact, purchasing has to secure the supply of goods, and logistics has to make sure the products can be delivered.

In a future organization, management could give an AI-based marketing department responsibility for large parts of the launch.

The AI department researches the market, identifies customer groups, drafts campaign proposals, creates content, and plans the advertising. At the same time, the agents work with the sales department to understand customers' needs, and with purchasing and logistics to check what the company can actually deliver.

If a campaign promises delivery within two days, but the logistics system shows that the product won't be available for three weeks, the agents could catch this before the campaign goes live.

The interesting part isn't that AI can write the ad copy. It's that AI can understand how marketing is connected to sales, purchasing, and logistics, and coordinate the work across departments.

Management doesn't necessarily need to follow up on every single task. It can give the AI department a goal, a budget, and some boundaries, and then evaluate the proposals and the results.

For now, a complete, independent department like this is still a future scenario. Parts of this way of working are already possible, but this requires access to the relevant systems, reliable data, clear authority, and control over what the agents do.

Why more advanced AI could be easier to adopt

One of the most interesting points in the article is that more advanced AI could actually be easier for companies to adopt than simpler AI tools.

That may sound backward, but the difference lies in who has to adapt to whom.

When a company starts using a new program, employees usually have to learn how it works. The company may have to change routines, provide training, and adapt existing processes to the software.

With a capable AI agent, the relationship could be different.

Imagine that a company hires an experienced logistics manager. This person is expected to get to know the company's systems, understand the suppliers, learn the routines, and figure out how the job is best done. The company doesn't need to explain every single action.

This is the ability the authors believe AI could gradually develop. Instead of the company having to learn the technology, the technology could increasingly learn the company.

A manager could then describe what needs to be achieved, while the AI agents work out for themselves how the task should be solved. They could ask questions, gather the information they need, suggest improvements, and coordinate the execution.

In this way, AI moves from being software that has to be operated to working more like a coworker who is given responsibility for a result.

Between us. When AI is placed between management and employees

The third scenario is perhaps the most unexpected. Here AI gets a role usually associated with management, coordination, and control.

We are used to employees reporting to middle managers, who in turn report further up the organization. But what happens if AI can keep track of large parts of the company's activities, understand how things are connected, and spot problems before the people do?

Picture a company with sales, purchasing, a warehouse, and logistics.

The sales department sees growing demand for a product and promises customers short delivery times. The purchasing department knows the supplier has capacity problems, while the logistics department is already seeing signs of delays.

The information exists in the company, but it is spread across different systems, departments, and employees. No one necessarily has the full picture.

An AI agent with access to the relevant information could spot the connection. It sees that the sales forecasts don't match the supplier's capacity, and that the company risks promising customers something it can't deliver.

Instead of waiting for the problem to be discovered through reports and meetings, the AI agent alerts management and suggests what to do.

It could, for example, recommend contacting an alternative supplier, adjusting the marketing, or changing the delivery times communicated to customers.

So the AI agent hasn't just automated a task. It has done part of the job that usually falls to a coordinator, middle manager, or advisor.

Could AI also become a kind of middle manager?

The authors believe that some management tasks may be better candidates for automation than we have traditionally imagined.

Much of management work consists of gathering information, understanding what is going on, weighing alternatives, prioritizing, and making sure different departments are pulling in the same direction. These are largely information-based tasks.

At the same time, practical work that requires physical presence, handling unpredictable situations, and direct human contact may be harder to automate.

So it is not a given that automation always starts at the bottom of the organization and gradually moves up. Some tasks at the management level could be automated before the tasks of the people they manage.

That doesn't necessarily mean AI becomes the employees' boss. A more natural development is that AI works as a continuous advisor to management, with an overview of the business and the ability to spot problems, opportunities, and anomalies.

A setup like this could also make it easier to see valuable contributions from employees that would otherwise go unnoticed. At the same time, it could be used for extensive surveillance and micromanagement.

That is why there is an important difference between using AI to understand how the company works and using it to control everything employees do. How much autonomy employees get will depend on how management chooses to use the technology, and what limits are set on how much management can see and control.

Fewer meetings, less planning, and faster decisions?

If AI can do more and more of the work, the need for traditional processes could also change.

Today, many companies spend a lot of time on meetings, project plans, approvals, and coordination before a new initiative can get started. One important reason is that mistakes are often expensive and time-consuming. When a project requires many employees and several months of work, management naturally wants to be sure it is worth doing.

But what if it becomes possible to test an idea in a few hours instead of several weeks?

A marketer who wants to explore a new market might not first need to write a detailed project description, get the budget approved, and wait for a development department to free up capacity.

They could use AI to build a working prototype, put together a campaign proposal, or do an initial market analysis. The results could then be presented to management.

It could become cheaper to try something in practice than to spend a long time discussing whether it should be tried.

This could reduce the need for some approval processes and make it easier for employees to take the initiative. At the same time, major financial decisions, privacy, security, and other risky matters will still require clear oversight.

And there is an opposite possibility. The same AI technology that makes employees more independent could also make it easier for management to keep track of and control their decisions.

That is why AI won't automatically lead to less bureaucracy. It can both reduce it and reinforce it.

Will the companies of the future be smaller or larger?

This is another central question in the article.

Traditionally, one of the advantages of large companies has been that they can bring many people and resources together under one organization. That makes it easier to coordinate production, share information, and carry out large projects.

But organizations like these also have challenges. The more employees and departments there are, the more time can go to communication, reporting, and coordination.

AI could change this in both directions.

On the one hand, small companies could become far more capable than before. A business with only a few employees could use AI agents to handle tasks that used to require separate departments for sales, marketing, customer service, and administration.

A small online store could, for example, have AI agents analyze demand, research supplier prices, suggest orders, produce marketing content, and handle parts of its communication with customers.

Then a small number of people could run a business with a capacity that used to be reserved for far larger organizations.

On the other hand, AI could also make large companies more efficient. When communication and coordination get easier, management could get a better overview of businesses with many departments, projects, and geographic locations.

It could also become easier for companies to work with outside suppliers.

When AI agents work together across companies

Imagine a retailer that buys products from several distributors.

Today, much of the communication goes through online stores, emails, order systems, and customer service. The buyer has to check prices and availability, place orders, and follow up on any problems.

In a more AI-based value chain, the retailer's AI purchasing agent could communicate directly with the distributors' systems or AI agents.

It could check which products are available, compare prices and delivery times, and suggest how an order should be split between different suppliers. If a shipment is delayed, the agents could exchange information and find alternative solutions within agreed limits.

People would only need to get involved when something comes up that requires a decision, a negotiation, or a judgment beyond the authority the agents have been given.

This could reduce the need for manual coordination between the companies. But it requires that the systems can communicate, that the information is reliable, and that there are clear agreements about responsibility and what the agents are allowed to do.

If cooperation between different companies becomes that easy, companies could also choose to outsource more tasks instead of handling everything in-house.

The result could therefore be both a greater number of small, specialized companies and larger businesses that coordinate many projects and suppliers with the help of AI.

What happens to skills, responsibility, and pay?

When AI can do more and more tasks, the skills that become valuable also change.

If two employees have access to roughly the same AI tools and both can get help with analysis, planning, and execution, it becomes harder to stand out just through the ability to produce content or process information.

Then other qualities could matter more.

These include the ability to understand what actually matters, set the right goals, assess the quality of the results, and make good decisions. Knowledge of the field will still be necessary, both to give AI the right tasks and to notice when it gets things wrong.

An experienced buyer knows, for example, that the cheapest supplier isn't necessarily the best. Delivery reliability, quality, payment terms, and past experience can be decisive. AI can analyze the information, but the judgments have to be reliable and suited to the company's needs.

When one person, with the help of AI, can be responsible for far greater assets or projects than before, the quality of that person's decisions could also have greater financial consequences.

The authors therefore point to two possible developments in pay gaps. If AI makes everyone better at making good decisions, the differences between employees could shrink. But if some people are significantly better at managing AI agents and making crucial decisions, the differences could also grow.

We don't yet know how this will play out.

It's no longer just about automating tasks

The overall message of The Future of the Firm is that AI could matter far more for companies than traditional efficiency gains.

In the past, we have largely used technology to make existing tasks faster. Spreadsheets made finance work easier, CRM systems (customer relationship management software) improved sales follow-up, and ERP systems (enterprise resource planning software) brought together information about things like purchasing, inventory, and finances.

AI can of course continue this development. But AI agents also open up another possibility, that the technology can carry out larger parts of a process, work across functions, and increasingly figure out how tasks should be solved.

That makes it natural to ask a different question than before.

Instead of just asking which tasks AI can automate, companies could start exploring which entire processes and areas of responsibility AI can handle, which decisions should still be made by people, and how people and agents can best work together.

This doesn't mean that every company will be run by AI agents, or that people will become redundant. Much of this is still a set of possible future scenarios, and the development will depend on how reliable the technology is, the economics, regulation, and how much responsibility we actually want to give AI.

But if AI eventually becomes able to handle ever larger parts of the business, the consequences won't only be felt in how we do the work.

They will also be felt in how we distribute responsibility, build departments, hire people, and organize the entire company.

And that could turn out to be a far bigger change than simply getting more done in less time.

Sources and further reading