Agents at Work Videos

Lightly edited for clarity.

AI agents don’t just generate, they execute. So think about how you use something like ChatGPT today. You ask it to write something, summarize something, explain something. It gives you an answer. Hopefully you iterate, but then you’re done.

An agent doesn’t stop. It’s the difference between someone who writes you an email explaining what you need to do and someone who writes you an email explaining about the work they’ve already done.

So here’s what that looks like in practice. Instead of asking AI to summarize a Jira ticket, you give an agent access to Jira and GitHub and say, pull everything that changed in the last 24 hours and generate a status update. The agent AI will query Jira, it’ll read the GitHub commit history, it’ll correlate those changes, and produce a report.

And that’s not one prompt. That’s a sequence of actions across real systems. And once you have that foundation, where agents can connect to your systems, then you can start to expand on them and leverage some advanced capabilities of AI.

So not just pull everything that changed in the last 24 hours and give me a report, but review the requirements for ticket AI X5, 78. Review the code that was created in branch bug fix AI X5, 78. Analyze what requirements were met, what requirements possibly weren’t met, and specific areas that should be focused on by manual QA.

The agents sit inside your workflows. They’ll take actions on your behalf, and when you implement them properly, they can dramatically extend the capabilities and efficiencies of your team.

Lightly edited for clarity.

If you want to understand where agents fit, look for the boring parts of the job. The best place to find agent use cases is not in high-level planning; it’s in coordination work. It’s the tedious, cross-system, nobody-wants-to-do-it layer that sits between decisions and results.

Take marketing teams. A typical weekly report is not really analysis. It’s pulling numbers from analytics tools, combining it with campaign data, and then formatting it into slides or dashboards.

Same in software development. Status reporting usually means checking tickets, and reviewing recent commits, and piecing together what was actually shipped. Things like bug fixing and ADA issues, SEO issues, browser specific things, they’re all straightforward, well defined tasks, but they’re tedious and time consuming for developers and cloud.

Agents are ideal for these cases, not just writing code, but also generating summaries and PR descriptions and workflow outputs.

And it’s not just in tech. In a law firm, agents can pull case history, flag relevant precedents, and draft a brief outline before a human attorney ever opens it.

And that pattern is the same everywhere. The work that connects systems is where agents win, not replacing the decision, just handling the legwork that surrounds it.

Lightly edited for clarity.

Agents don’t replace people, they change what people are responsible for. If agents actually replace people, companies would already be running them into end to end. But that’s not what I see happening.

So let’s talk about how they’re actually being used. Take coding workflows, for example. An agent can take an issue, generate code, and open a pull request, but it doesn’t merge it into production. A human reviews it. Why? Because the hard part’s not writing the code, it’s knowing whether it’s the right code.

Does it break something else? Does it fit the architecture? Does it handle edge cases? Does it actually solve what the customer need is? And agents don’t necessarily have that full context.

So what happens is execution gets faster, but the responsibility shifts. You’ll spend less time doing the work, but more time evaluating whether that work is correct.

And that part is something most people underestimate. It’s not the job that disappears, it’s what you’re responsible for that changes.