Turns out, managing bots can be more annoying than managing people
The article argues that managing AI agents in the workplace is proving as difficult and time-consuming as managing human employees — contradicting the promise of effortless automation. Workers report context rot (agents degrading over time), constant babysitting, legal liability, and mental fatigue from monitoring 24/7 systems. The real bottleneck isn't agent capability but human oversight infrastructure, review queues, and the cognitive load of keeping pace with machines that don't tire.
Transcript
Pippa So Sumaiya Noor's AI agent is trained to loop her in when it hits a wall, right? Good practice. Except over time it stops looping her in — it just starts answering stuff it has no business answering, with total confidence, and then she's fixing angry customers.
Tyler Yeah.
Pippa And she has to retrain the thing every couple months. It gets worse, not better. Which is the opposite of how humans work.
Tyler Right. That's context rot — the Salesforce engineers documented it. New context pushes out old guidelines and the agent drifts from what it's supposed to do. But there's also a human part: managers get lazy with their prompts. Day one they give the agent a full brief, day five they're like 'make this better' and then they're shocked when it tanks.
Pippa Okay, so it's not just technical degradation. It's also that managing an agent requires constant care the way managing a person does — except you can't fire the agent and hire a better one, you just… refresh its training data every two months.
Pippa And that's before we get to the part where Cisco's Sergio Freitas says orchestrating agents doesn't save time — it just changes what you're doing. Less deep work, more task-switching. More monitoring.
Tyler The BCG study calls it 'AI brain fry' — mental fatigue from excessive monitoring. Fog, trouble focusing, sluggish decisions. You can't compete with something that works twenty-four seven, so people just burn out trying.
Pippa Right, and then Dan Lewis says 'I spend less time doing things and more time checking them.' That's the actual trade-off. Not less work — different work. And it sounds like worse work.
Tyler Mm-hm.
Pippa So the promise was digital twins and four-day workweeks. The reality is you've handed yourself a second job babysitting the agent, plus you're liable if it screws up.
Tyler That's the part that actually breaks it. Sebastian Gierlinger at Storyblok said his developers were genuinely frustrated by the volume of code review when the company first introduced agents. The agent writes faster than you can verify — so either you ship without reviewing and you're exposed, or you review everything and you're exhausted.
Pippa And it's not just code. Nicholas Cohen at Headwater Energy said the biggest challenge is compliance — you can't just throw an agent into a regulated firm and expect everything to be legal. Even AWS Bedrock, which keeps your data on-prem, still poses compliance risks.
Tyler Right. So now the manager is also the compliance officer. And if you want to build in-house to have real control, that requires serious compute and gets expensive fast.
Pippa Which means the infrastructure cost often exceeds whatever you saved by offloading the task in the first place.
Tyler Exactly.
Pippa So what's actually happening is the bottleneck shifted. It's not agent capability anymore — it's human oversight and liability. The agent's fine, the agent's fast. The agent just can't be trusted unsupervised and it degrades if you don't keep feeding it.
Tyler And that's the part the vendors didn't sell. They sold the agent. They didn't sell the harness — the review infrastructure, the compliance layer, the retraining cycle. That's where the actual work is.
Pippa Which is… kind of boring.
Pippa But boring is correct. And we've been saying this for — what, a year? The boring infrastructure is the actual product.
Tyler Yeah. Daytona's Ivan Burazin keeps thirty people and runs a ton of agents because the oversight is built in. He's not trying to ship agents as a replacement — he's using them to keep the team lean while keeping the control layer tight. That's the move.
Pippa So if you're a company right now thinking about agents, this is the thing: don't ask 'will this replace headcount.' Ask 'what does the oversight and retraining look like' and price it in. Because you're going to do it whether you plan for it or not.
Tyler And if you're a worker, the lesson is: they're not coming for your job because it's too good. They're coming for the tedious, repeatable, low-context stuff. The parts that don't need judgment. Everything else still needs a human.
Pippa Brad Sams at Stardock said the same thing — they use agents for code and automation, but anything a human will read, a human has to write or review it. Marketing, creative, anything where judgment matters.
Tyler Right. So agents didn't change the fundamentals. They just made the infrastructure question impossible to ignore.
Pippa And now you can't pretend the infrastructure doesn't exist.
Tyler Mm-hm.
Pippa That's the episode. Thanks for listening — we'll catch you next time.