Task Complexity Analyzer

Describe a task you're considering automating and the analyzer scores it on three axes: complexity, repeatability, and tolerance for error. The combined score tells you whether the task is a good fit for an AI agent, a simple script, or a human review pass.

Good automation candidates score high on repeatability and low on error-cost — think classification, summarization, and data extraction. Poor candidates have legal exposure, require live judgment calls, or change format every week.

The scoring is heuristic, not predictive. Run a small pilot before scaling anything the analyzer flags as borderline.

How the score is built: each axis gets 1-5. Repeatability rewards tasks that look similar every time (a 5 = identical structure, only the data changes). Complexity penalizes tasks that need multi-step reasoning or external context (a 5 = high). Error tolerance asks how bad a wrong output is (a 5 = trivial to spot and fix; a 1 = ships to a customer or a regulator). The composite leans repeatability heavy — the same task done a thousand times pays back automation faster than a one-off masterpiece.

Worked example: "summarize 200 customer support emails per day into a daily digest." Repeatability: 5 (same shape every time). Complexity: 2 (just summarization). Error tolerance: 4 (humans skim the digest; bad summaries get caught). Score lands in the green. Compare to "draft legally binding contracts from a template." Repeatability: 4. Complexity: 4. Error tolerance: 1 — a wrong clause costs real money. Score lands red. Pilot the first; don't pilot the second without a lawyer in the loop.

Where the heuristic breaks: tasks with hidden complexity. "Tag these images" sounds simple until you discover the cohort has 47 edge cases the labels don't cover. The analyzer can't see what you don't tell it. Run 20-50 real examples through your candidate solution and measure agreement with a human gold standard before committing. If accuracy is below 90%, the task is harder than you described — go re-score with the truth.

A practical scoring rubric: complexity 1 (single-step, well-bounded) to 5 (multi-stage with branching logic). Repeatability 1 (one-off) to 5 (identical structure thousands of times). Error tolerance 1 (catastrophic if wrong) to 5 (trivial to spot and fix). The composite favors the 5-5-5 corner — high repeatability, low complexity, high error tolerance — which is the textbook automation candidate. Anything in the 1-1-1 corner stays human. Most real tasks land in the middle and need pilot testing to know whether they're worth the engineering investment.