LGLocal Gradient
Local GradientField note 001July 2026

A personal forecasting experiment

Estimating agent work in agent time.

I take the human-hour estimate a coding agent already gives me and translate it into agent labor and wall-clock time using evidence from my own traces.

A misty mountain trail dividing at a wooden distance-and-hours sign
One trail. Several honest ways to describe the distance.

The agent has already done useful planning when it says “roughly 75 to 130 hours.” I do not want to throw that estimate away. I want to change its units.

My forward skill takes the agent's own human-hour estimate as its starting point, converts it with coefficients derived from completed session traces, then accounts for dependencies and parallel work. The look-back is supporting calibration: it helps the conversion improve after each run.

01

The central conversion

Keep the estimate. Change the units.

What the agent gives youHuman-equivalent effort

Work size for one experienced person working alone. Keep the agent's own breakdown, range, assumptions and dependencies.

÷ 43.3

Your own trace-derived coefficient — human-equivalent hours per attended agent hour.

What the work costsAgent labor

The sum of every worker's time. Four agents running for 15 minutes is about one hour of agent labor.

→ critical path

Dependencies, real parallelism and the serial gates you cannot skip.

What you actually waitAttended wall-clock

Elapsed time from dispatch to a verified stopping point, including serial integration.

02

Paying it forward

Take the skills—and the coefficients.

Todd shared the seed in public, so I want to do the same with my fork. These are the complete, portable Markdown skills I currently use—not abbreviated prompts or screenshots. Each file includes the real 43.3× conversion, the 60/35/22/14/10 lines-per-hour tiers, the separate 120-lines-per-hour cleanup rate, formulas, guardrails and output templates. Copy one into a SKILL.md file wherever your agent loads skills, then replace my starting coefficients as your own actuals accumulate.

Before the work

Forward estimation

  1. Ask the agent for its conventional human-hour estimate.
  2. Preserve the task breakdown, range and dependencies.
  3. Cross-check the input with likely files and complexity.
  4. Convert each item to agent labor with trace-derived calibration.
  5. Model parallel work as a critical path, not a naïve division.
  6. Report the range, basis and uncertainty.
Download Markdown

After the work

Look-back accounting

  1. Inspect the completed git range.
  2. Exclude generated and vendored material.
  3. Classify insertions by file complexity.
  4. Keep deletion-only cleanup separate.
  5. Record measured wall-clock and agent labor when available.
  6. Append an anonymized row to the private ledger.
Download Markdown

Current personal coefficients

43.3×

Human-equivalent hours per attended agent hour, derived from 243 anonymized work-item rows over four weeks.

  • T1TrivialConfiguration and boilerplate60 lines/hr
  • T2RoutineTests, documentation, simple UI and scripts35 lines/hr
  • T3MediumBusiness logic, data access and integrations22 lines/hr
  • T4HighArchitecture, migrations and state machines14 lines/hr
  • T5SpecialistParsers, systems, cryptography and GPU work10 lines/hr
  • ΔCleanupDeletion-only credit, kept separate120 lines/hr
03

Good enough to prioritize

From 64 hours to 1.5 was the useful result.

Agent estimate
75–130 human hours
Estimate from cost-estimate skill
≈64 human-equivalent hours≈1.5 total agent-labor hours
Estimate from look-back skill
59.7 human-equivalent hours
Wall-clock time
≈7 minutesObserved with the Grok 4.5 harness

The human-equivalent work estimate held up: about 64 hours going in versus 59.7 in the look-back.

Neither the original agent estimate nor my conversion assumed parallel execution. The ≈1.5-hour result is total agent labor, not a parallel wall-clock forecast.

The observed run used Grok 4.5. In my measurements, its prefill and decode speeds were at least 50% faster than the Codex and Claude API routes behind my coefficient, and the harness also ran work in parallel. Those unmodeled advantages help explain the ≈7-minute wall-clock.

I do not plan to model every one of those variables. Going from about 64 human-equivalent hours to about 1.5 agent-labor hours is enough for me to prioritize what can run while I walk my dog.

04

Credit before conclusions

The look-back idea did not originate with me.

This work was seeded by Todd Saunders’s public /cost-estimate prompt: lines of code plus complexity translated into senior-developer hours. I used it publicly on March 5, 2026 with both hours and dollar return-on-investment framing.

My current fork retires the dollar layer, keeps hours-only look-back accounting, separates agent labor from attended wall-clock, and calibrates a forward forecast against my own session ledger. Julien Barbier later built a parallel expansion of Todd’s idea; it was not the direct source of my fork.