How to measure the human cost of agent work

#364 – November 15, 2026

a method for tracking time spent on supervision, reviews, corrections, and verification, alongside agent expenses

How to measure the human cost of agent work
7 minutes by Michael Cutler

Michael explains how to measure the real human effort and costs of using AI agents. He introduces a method for tracking time spent on supervision, reviews, corrections, and verification, alongside agent expenses. He also shows how to account for failed attempts and manual work to understand whether AI agents truly improve productivity and save time.

A Pile of MCP Connectors Is Not a Context Engine
sponsored by Unblocked

Your team connected Slack, Jira and GitHub over MCP, and engineers still spend their time steering agents to the right answer. This article looks at why: connectors fetch from one tool at a time, so every agent stitches the pieces together itself, trusts the first doc that looks right, and adds a search to your token bill each time. Read what changes when a context engine sits on top of your connectors: the same coding task finished 83% faster at 48% lower cost.

The tea tasters
8 minutes by Mike Fisher

Mike explores how organizations sometimes keep problems alive because entire departments depend on them for their roles and resources. Using real examples, he explains how this can lead to unnecessary processes and resistance to change. He also offers practical advice for leaders on how to reward problem prevention, encourage teams to take on new challenges, and recognize when a department’s original purpose has been fulfilled.

Business continuity: From Black Friday to semester start
6 minutes by Kore Nordmann

Good backup and recovery practices apply everywhere, but what counts as "critical" depends on context. Kore argues for a university learning analytics platform, student consent data matters far more than uptime, since a GDPR violation is a worse outcome than a slow dashboard. The platform is also designed from day one for handover to other operators, with short runbooks and drills that the receiving team must complete before taking responsibility.

Predictability vs outcomes
4 minutes by Marty Cagan

AI is making it easier than ever to ship features, but actual results remain scarce. The real problem is not that teams focus on output over outcomes. It is that most companies secretly prioritize predictability over outcomes, hoping good results will follow from shipping what was promised. Achieving real outcomes requires admitting uncertainty and testing ideas before committing to them, which predictability-focused cultures resist.

The loss of positive friction
3 minutes by John Cutler

According to John, constraints that slow us down often also force better decisions, and removing them removes both the bad and the good. The friction in old workflows, like waiting on approvals, pairing sessions, or rounds of doc revisions, created space to catch mistakes and ask hard questions. AI tools now remove that friction fast, but also remove those moments. We need to be deliberate about building them back in.

How to give your agents the right context, fast enough, without starting from scratch
sponsored by Redis

Are your AI agents costing more in tokens than they return? Few teams see repeatable production success with agents, and the ones that work often burn tokens on over-stuffed prompts, redundant LLM calls, and brittle integrations. Redis CEO Rowan Trollope explains why agents are software, and how Redis Iris gives them the right context, fast, so they work in production from day one.

industry

security

And the most popular article from the last issue was:

newsletters