Agents and the illusion of productivity

I recently released a new intensive course on Grafana MCP. It is 2 hours long, with 30 videos across 7 chapters. Making the course took 2-3 months of work that I mostly did alone. But I had help — AI, which led me into a trap.

You see, because AI is so good at doing several tasks now, it’s easy to think you can save time and money by orchestrating several agents to do things you’d have normally outsourced. Bridging several domains with AI gives you an illusion of productivity, but it can hurt you.

I’ve made three courses now and the creative process is straightforward. Research → Write scripts → Plan labs → Implement labs → Record → Edit → Review and correct → Release → Marketing.

Each of these steps requires careful attention to detail because my students need to get the right information that sometimes changes their careers. And each of these can be assisted by AI one way or another.

My flow before AI was exactly the same, but I NEVER handled editing and review for a course. I always had a LinkedIn Learning producer and editorial team behind me. But being my first independent course, I decided to see how I could use agents for editing and review. This was a mistake.

If you’ve read 10x Is Easier Than 2x, you’d be familiar with the idea that you can make outsized gains in any endeavour by focusing on the 20% of the effort that gets you 10x gains instead of the 80% that gets you 2x. The 80% is often the mundane work like fact-checking, while the 20% is improving your odds by cold-emailing or doing market research.

With agents now ready and willing to handle all sorts of tasks, it’s easy to picture yourself as a ship captain orchestrating tasks and only focusing on the 20% of the work that brings you 10x gains. This holds true for most thinly-scoped tasks within a specific job role, but not for a joint production like a course, which requires different roles and methods of thinking.

If I hadn’t taken on that 80%, the extra time would’ve gone to the work in green. Instead, it went to the work in red.

The illusion manifests itself in two ways. First, while you get those extra tasks done quicker, you are now responsible for all the context that would’ve been in other people’s heads. And second, you also bear the burden of quality control for each of those tasks. This means you’ll send a prompt, wait, review and make corrections. A human working in such a different role would’ve only given you the context relevant to your role, and ensured a level of quality before presenting to you.

Ultimately, the course was completed and released, but not without costing me a lot of other work that would’ve otherwise gone into getting a 10x greater chance of success. It also came at a cost of stress, and lots of it. I can’t scale the amount and quality of courses I make if it costs so much physically and mentally to make one.

When making the next course, I’ll still use agents extensively for research and general assistance, but I’ll be focusing on the 10x.