HumanHoursHUMANHOURS
HUMANHOURS / BLOG

Every agent, measured in human hours.

Notes on the product, the engineering, and the math behind agent ROI. Written for the people building, shipping, and paying for AI agents.

RSS

One short note when we publish. Unsubscribe any time.

3 AUGUST 2026 · 5 MIN

The METR AI study: devs said 20% faster, the clock said 19% slower

METR's randomized trial found developers who felt 20% faster with AI were measured 19% slower. Self-reported time savings need a measured baseline.

  • METRICS
READ POST 30 JULY 2026 · 4 MIN

Agentic AI value tracking per run: chasing is easy, catching is not

Three quarters of enterprise leaders say they are adopting agentic AI. Few can show the value. Per-run tracking of hours and money saved fixes that.

  • AGENTS
  • METRICS
READ POST 27 JULY 2026 · 4 MIN

Who owns AI ROI reporting? Usually nobody

Engineering owns traces, finance owns budgets, and hours saved falls in the gap. Why AI ROI reporting needs one named owner and one metric.

  • COMPANY
  • METRICS
READ POST 23 JULY 2026 · 5 MIN

How to measure AI agent performance when every app ships agents

Gartner expects 40% of enterprise apps to ship task-specific agents by end of 2026. Almost none report task value. How to instrument it from day one.

  • AGENTS
  • METRICS
READ POST 20 JULY 2026 · 4 MIN

Prove AI agent ROI to finance before 25% of budgets slip to 2027

Forrester expects enterprises to defer a quarter of planned AI spend to 2027 because ROI stayed unproven. The per-task metric that keeps your budget funded.

  • METRICS
  • COMPANY
READ POST 13 JULY 2026 · 4 MIN

AI workslop eats up to 38% of your reported time savings

Forrester puts unmeasured rework at 22 to 38 percent of self-reported AI time savings. Net out the workslop or the number is fiction.

  • METRICS
READ POST 9 JULY 2026 · 5 MIN

The agent autonomy ratio: one number that shows trust is growing

Agent autonomy should climb over time. This metric turns uptime and human wait time into one number that shows whether an agent is maturing or just running.

  • PRODUCT
  • METRICS
READ POST 6 JULY 2026 · 4 MIN

Fixed vs dynamic: getting the n8n time saved calculation right

n8n reports hours saved only as accurately as you wire it. When to use fixed vs dynamic time saved, and how to keep the number defensible.

  • ENGINEERING
  • METRICS
READ POST 29 JUNE 2026 · 4 MIN

Why agentic AI projects fail, and the bottleneck behind the 40%

Gartner expects over 40% of agentic AI projects canceled by 2027. The frontier model is rarely the problem. The bottleneck is measurement.

  • AGENTS
  • METRICS
READ POST 25 JUNE 2026 · 4 MIN

Tie AI value to the P&L before the 2027 budget cut

Forrester says a quarter of AI budgets slip to 2027 and fewer than 1 in 3 leaders can tie AI value to the P&L. Instrument value from day one.

  • COMPANY
  • METRICS
READ POST 22 JUNE 2026 · 5 MIN

The fully loaded AI agent cost your token bill hides

Token spend is often a minority of the true AI agent cost per task. Here is how to build a fully loaded number that survives a CFO review.

  • METRICS
  • ENGINEERING
READ POST 18 JUNE 2026 · 4 MIN

AI compliance monitoring when data cannot leave your environment

Regulated teams cannot ship AI payloads to a third-party tracker. How to run AI compliance monitoring and prove hours saved with metadata only.

  • ENGINEERING
  • COMPANY
READ POST 15 JUNE 2026 · 4 MIN

The hidden cost of untracked AI agents

Untracked automations get cut first when budgets tighten. The three hidden costs of running AI agents, and the minimum tracking that fixes each.

  • COMPANY
READ POST 11 JUNE 2026 · 4 MIN

Human-equivalent hours saved: the metric that survives a CFO review

What human-equivalent hours are, the three calibration steps that make them defensible, and a worked example from raw executions to money saved.

  • METRICS
  • COMPANY
READ POST 8 JUNE 2026 · 3 MIN

Built-in task types vs custom: when to break out of the vocabulary

HumanHours ships a vocabulary of common task types. The rule for when to override a baseline, when to build a custom type, and when to leave the built-in alone.

  • PRODUCT
  • METRICS
READ POST 4 JUNE 2026 · 5 MIN

Per-execution vs per-record tracking: when each makes sense

A flow that runs hourly but acts on 3 records is a different tracking problem than a webhook that fires once per record. The decision rule, the failure modes, and what each does to your dashboard.

  • ENGINEERING
  • METRICS
READ POST 1 JUNE 2026 · 5 MIN

The 11-by-11 tipping point: when AI savings start compounding

Microsoft found that 11 minutes saved per day over 11 weeks is the threshold where AI shifts from novelty to habit. The leading indicator that tells you which agents are on track to cross it, and which to cut.

  • AGENTS
  • METRICS
READ POST 28 MAY 2026 · 3 MIN

Human-in-the-loop ROI: only count what actually replaced human work

An agent drafts 800 reply suggestions. Humans approve 240. The dashboard reads 800. Place the tracking node after the approval gate, not before, and the number stops inflating.

  • METRICS
  • AGENTS
READ POST 25 MAY 2026 · 5 MIN

Counting AI agent tool calls: per call or per parent execution

An agent fires five tool calls to finish one task. Log five savings events or one? Neither default is right. Count what the human alternative replaced.

  • ENGINEERING
  • AGENTS
READ POST 21 MAY 2026 · 5 MIN

The 30/60/90 dashboard: a CFO-ready AI ROI scorecard

Adoption at 30 days, operational gains at 90, a P&L line at 12 months. The three artefacts an AI program needs to produce at each horizon to survive a finance review.

  • METRICS
  • COMPANY
READ POST 18 MAY 2026 · 5 MIN

n8n vs Make vs Zapier: which one tracks AI savings cleanly

n8n ships a built in time saved metric on Pro plus. Make and Zapier count operations and tasks. None roll up a CFO ready human hours figure.

  • ENGINEERING
  • PRODUCT
READ POST 17 MAY 2026 · 5 MIN

AI customer support ROI: $0.99 per AI ticket vs $6 per human one

AI tickets cost about a dollar. Human tickets cost six to fifteen. The 210% AI support ROI headline is real, but the unit economics behind it rarely show up in the pitch.

  • AGENTS
  • METRICS
READ POST 11 MAY 2026 · 5 MIN

How to set an AI productivity baseline before you deploy your first agent

Three ways to calibrate an AI productivity baseline that holds up under finance review: time-tracked sample, expert consensus, historical logs.

  • METRICS
READ POST 10 MAY 2026 · 4 MIN

Why most GenAI projects fail to prove AI ROI

MIT found 95 percent of GenAI pilots show no measurable return. The diagnosis is wrong: AI is not failing on the model side, it is failing on measurement.

  • COMPANY
  • METRICS
READ POST 9 MAY 2026 · 6 MIN

Where to track AI hours saved: a decision tree for any automation stack

Eight automation patterns, eight different places to drop the tracking event. A decision tree for keeping AI hours saved defensible across n8n, Zapier and custom Python.

  • ENGINEERING
READ POST