From Chat Window to Coworker: How Far Practical AI Has Come

Every few weeks there is another AI announcement, and it is easy to let them blur together. Let’s reflect on how far practical, everyday AI has actually come, and where it is heading next.

Phase 1 – Yesterday: AI you talk to.

The first phase is the one most people know. You open a chat window, ask a question, and get an answer. Over time it became more useful: you could upload a document and get a summary, download a finished draft, and connect the AI to your email, calendar, or shared drive so it could look things up for you.

That was a real step forward, but you were still the engine. Every task started with you typing a prompt, and every result ended with you copying, pasting, and saving it somewhere. The AI was a very smart assistant sitting in a separate window.

Phase 2 – Today: AI that works alongside you.

This is where many organizations are right now, and where tools like Claude Cowork come in. Instead of passing files back and forth, the AI works directly with your files and folders. Instead of answering one question, it carries out multistep work : reconciling a spreadsheet, building a board packet from last quarter’s materials, filing documents into the right folders based on your processes.

And it does not have to wait for you. Tasks can be scheduled, so the weekly report, the Monday briefing, or the month-end checklist runs on its own and is ready when you are.

For a lean nonprofit or professional services team, this is where the time savings stop being anecdotal and start showing up on the calendar.

Phase 3 – Tomorrow: AI that is always on.

The next phase is already arriving, and it flips the relationship. In just the last month, OpenAI introduced Dots and Anthropic introduced Claude Duet, both early versions of always-on agents. Until now, every interaction with AI started with you. These agents start the conversation themselves. They keep an eye on your connected apps, keep working toward a goal after you log off, and reach out when something needs your attention, whether that is a morning briefing or a message in Slack or Teams. Picture starting your day with a briefing that already flags the grant or proposal deadline, the unanswered email from a critical stakeholder, and the vendor invoice that does not match the contract, with drafts ready for your review.

These tools are still early previews, and for now they are only available on certain plans. Both companies are candid that the agents can make mistakes, and both keep a person in the loop for sensitive or consequential actions.

Across the three phases, the question shifts:

• From “What can I ask it?” to “What can I hand off?” to “What should it be watching for?”
• From one person and one prompt to a process that runs on its own
• From reviewing answers to setting boundaries

As AI takes on more work independently, organizations need clear rules about what it can access, which actions require approval, and who checks its work. Those rules need to be in place before adopting phase three, when AI agents work continuously on their own.

For now, focus on phase two. Use it to learn where AI delivers real value, build your team’s confidence, and put appropriate permissions and oversight in place. That experience will prepare your organization to adopt more autonomous AI as the technology matures.

It has been a remarkable few years. The next few will move even faster. Contact us at insource@insourceservices.com to learn how we can help.