AI Can Now Use Your Software Like an Employee

For years, automating business software has required software to cooperate. Developers connect APIs, build integrations, configure webhooks, or create custom connectors that let one system talk to another. When an application lacks those interfaces, automation gets expensive quickly.

AI agents are starting to offer another way in.

Last week, OpenAI released GPT-6 Astra, a new model with much stronger computer-use capabilities. Astra can navigate browsers, spreadsheets, websites, and desktop applications, create documents and presentations, and complete multistep workflows through the interfaces people already use.

For technology companies, the interesting part of the release has little to do with OpenAI's claims about entering an "AGI era." Computer use could change how companies connect AI to the enormous amount of software that was never designed for agents.

Most enterprise AI systems depend on some combination of APIs, connectors, plugins, MCP servers, and custom integrations. Those connections give agents structured ways to retrieve information and take actions inside other applications.

They also require engineering work. Every application needs to expose the right functionality, and someone has to build and maintain the connection.

Computer-use agents approach the problem through an interface that almost every piece of business software already has: the one designed for people.

If an employee can open an application, find a customer record, enter information into a form, export a report, and move that information somewhere else, a capable computer-use agent can increasingly attempt the same workflow.

That matters most for software sitting outside the modern API ecosystem.

Large companies run on a mix of modern SaaS products and software that may be years or decades old. Internal tools, desktop applications, industry-specific systems, and older enterprise platforms often have limited APIs or none at all.

Automating workflows across those systems usually means building custom integrations, using brittle robotic process automation, or replacing the underlying software. Each option adds cost and can turn a seemingly simple automation project into a much larger engineering effort.

A reliable computer-use agent could work with some of those systems through their existing interfaces. A finance workflow might move between a modern cloud application, an old desktop accounting system, a spreadsheet, and a web portal without requiring a purpose-built integration for every step.

That could make automation viable in parts of the business where integration costs previously killed the economics.

Integrations are one of the hidden costs of enterprise AI.

An agent might be perfectly capable of deciding what needs to happen, but it still needs a way to interact with the systems where the work happens. Every new application can introduce another API, authentication method, connector, permission model, and maintenance burden.

Computer use gives teams another option. Some workflows may be easier to automate by teaching an agent to operate the application directly.

That does not make APIs obsolete. Structured integrations remain faster, more predictable, and easier to monitor for many high-volume or consequential workflows. Computer use expands the set of software that agents can reach, especially when a clean integration is unavailable or too expensive to justify.

For CTOs, this could change the calculation behind automation projects. The cost of connecting an agent to an old system may fall enough to make previously ignored workflows worth revisiting.

Most integration projects are organized around applications. Teams decide which systems need to connect, study their APIs, and build the infrastructure between them.

Computer-use agents make it easier to start with the workflow itself.

Imagine an operations employee who receives information by email, checks it against a customer record, enters something into an internal application, updates a spreadsheet, and sends a confirmation through another system. Today, automating that process could require several integrations.

An agent capable of navigating those tools can potentially execute more of the workflow as a single task. The engineering challenge becomes defining the outcome, giving the agent appropriate access, and deciding where human approval is required.

This is especially relevant for companies with years of accumulated software. They may be able to automate more of their existing operations without modernizing every underlying system first.

This issue of Redeployed is brought to you by Tecla Labs: As AI agents become capable of operating the same software employees use every day, companies need engineers who can turn those capabilities into reliable production workflows. That means understanding AI systems alongside APIs, permissions, legacy applications, cloud infrastructure, and the business processes connecting them. Tecla Labs helps companies turn AI into working business systems, starting with one high-value workflow, mapping how it really runs, and building a first version you can operate and improve. Book a free AI Assessment to find the workflow worth starting with.

APIs are designed for software. User interfaces are designed for people, which makes them a less predictable foundation for automation.

Buttons move. Page layouts change. Pop-ups appear. Applications load slowly. A field may be renamed or a workflow redesigned without warning. An agent operating visually has to recognize those changes and recover when the environment differs from what it expected.

Observability also gets harder. With an API integration, teams can log requests, validate structured responses, and restrict specific actions. An agent using a computer may navigate several screens and applications before completing a task, creating a much longer sequence of actions to monitor.

Permissions become especially important. Giving an agent access to a user interface can expose every action available to that user. Companies will need controls that determine what the agent can do, which actions require approval, and how its activity is recorded.

Computer use adds another layer to the architecture decisions behind enterprise agents.

Teams will need to decide when an agent should use an API, when it should operate an interface, and when a workflow deserves a dedicated integration. High-volume transactions may still justify direct connections, while lower-volume processes involving legacy software could be good candidates for computer use.

Reliability will shape those choices. Teams will need ways to test workflows across interface changes, detect when an agent gets stuck, capture its actions, and route failures to people. The engineering work moves beyond connecting systems and into supervising how agents operate across them.

This also creates an opportunity to revisit automation backlogs. Processes dismissed because a system lacked an API or the integration work was too expensive may now have another path.

For decades, APIs have determined which parts of software could easily participate in automated workflows. Computer-use agents loosen that constraint by giving AI access to the same interface companies already built for employees.

If the technology becomes reliable enough, the impact could be largest in the least modern parts of the enterprise. Old desktop software, internal tools, industry-specific applications, and awkward web portals suddenly become more accessible to automation.

Companies may discover that modernizing a workflow does not always require modernizing every system behind it. Sometimes the fastest path will be an agent that can simply use the software already there.

That makes computer use one of the more practical developments in the agent market. It gives AI a potential compatibility layer for the enormous installed base of software the enterprise has spent decades accumulating.

Connect With Other Technology Leaders

If you want to connect with other technology leaders having real conversations about AI and how it is changing business, check out GILD Curated Circuit.

More to come…

✔️ AI Is Disrupting Software Companies, but Not as Fast as Many Feared — The Wall Street Journal

✔️ Google Cloud, Accenture Launch Unit to Put AI Engineers On-Site With Customers — The Wall Street Journal

✔️ OpenAI's next big AI model has 'entered the AGI era' — The Verge