This as-told-to essay is based on a conversation with Sergio Freitas, senior director of engineering at Cisco, based in Lisbon, Portugal. His identity and employment have been verified by Business Insider. This story has been edited for length and clarity.
I started as a software engineer, became a manager, and currently work as a director of engineering at Cisco, with four teams based in Portugal.
Software engineering has always gone through changes. There are always new technologies you have to learn, like programming languages or frameworks. Now, it’s the way engineers work that’s changing rapidly, and in the last year, it’s changed significantly. People have to adapt to that, and that can cause stress.
The biggest change came after frontier models launched at the end of last year, like Anthropic’s Claude Opus 4.5. That’s when AI really started to be embedded in our workflows, and we began delegating and working with agents to do software engineering. We were using AI before, but more for autocomplete and prompting.
Agentic AI has changed our work
Our work as software engineers has changed from crafting the code to be focused on architecture, design, managing workflows and agentic orchestration. Engineers are shifting from manual work to becoming system architects and workflow orchestrators, allowing them to solve larger, more complex architectural problems.
The agents are like teammates and I manage 10 to 20 agents daily, which manage other agents. The number of agents they manage depends on the complexity of the task. The main orchestrator agent can spawn sub-agents to complete specific tasks.
They’re always running and they’re aware of collaboration platform notifications, GitHub pull requests, and other integrations when given explicit permission. For example, based on the description of a ticket in Jira, which is our project management and issue-tracking tool, an agent creates the code, and then that has to be reviewed by the person who owns the agent. It then goes through a review process, where at least two engineers have to review the code.
About four months ago, we moved into loop engineering, where, instead of the agent asking questions, it loops, or keeps working until it gets the final response.
Sometimes, depending on what needs to be done, the agent can ask for authorization to connect to certain systems or perform certain operations. More and more, it’s autonomous, but strict guardrails remain.
I’m reviewing more work
There’s a lot of value in having AI work with engineers, but they need to be conscious about reviewing the code. Everything that gets delivered to production is the human’s responsibility, not the agent’s.
I’m not working more hours, but there is an aspect of AI that brings, in many occasions, more work to the table because it can do more.
Cisco focuses a lot on security. So there are a lot of security scans you need to run, and conversations with the legal team that need to happen.
It’s a faster-paced job
As engineers adapt to this new way of working, some initial fatigue comes from the ability to do more. You have to shift your attention a lot, which slows you down if you’re not completing tasks end-to-end.
As engineers adapt to higher daily output and multi-threaded collaboration with agents, they need to build new mental habits around multitasking and asynchronous review, which is a different cognitive exercise from traditional coding.
It’s a wider scope. At the end of the day, you feel you were more focused, but not in the way when you’re coding and have to be in the zone for extended periods.
Crafting code was a skill that you built, and not many people had it. It kind of felt like a superpower of being able to create something from scratch and code it yourself. If you’re really into the code-crafting piece, it’s a big change that some engineers struggle with.
At the end of the day, it felt lighter. Now with the management of agents, it’s faster-paced.
Read the full article here
