CUSTOMER ACCOUNT / ATTRIBUTED EVIDENCE
What this case establishes
This is a commercial publisher’s customer account. Roseram checked the source and wrote the analysis; it did not interview the customer, run the product, audit the figures or establish that AI caused the reported business result. The pilot below is a proposal, not a completed experiment.
Primary material reviewed 2026-10-12. Analysis and proposed evaluation by Roseram Editorial.- 01Task acceptance criteria
- 02Budgeted model route
- 03Verified change and total cost
LegalOn: the documented workflow
OpenAI’s October 2026 account describes LegalOn routing coding work by complexity and applying team guidance and usage limits. The article’s headline, estimated daily reduction and mature-area reduction use different scopes; they should not be combined into a single universal savings percentage. Roseram did not audit billing or delivered code.
The mechanism worth examining
A small documentation change and a cross-service refactor impose different reasoning and verification requirements. Routing can reduce waste when the team knows which tasks a cheaper route completes acceptably. Token totals alone do not reveal whether the resulting change passes review. A low-cost attempt that needs three repairs can be more expensive than a stronger first attempt.
The implementation decision
Build a small set of representative tasks with known acceptance criteria. Route low-complexity changes first, while retaining an escalation path for unresolved failures. Apply account limits and per-task budgets that the team can inspect. Keep the experiment in a disposable branch; the case concerns engineering economics, not using a model to make unsupervised legal decisions.
Where the workflow can fail
A budget rule can truncate the useful part of an investigation or encourage a superficial patch. A high-capability route may also spend excessively searching unrelated files. Capture task context, route changes, tests and reviewer corrections. Distinguish a failed attempt from accepted output rather than counting every generated diff as completed productivity.
A practical evaluation, designed for this use case
Compare total request cost, runtime and human review minutes for the same task set. Include retries and failed attempts in the denominator. Inspect regression rates and whether developers can understand the result. The decision is whether a route meets the task’s acceptance bar at a reasonable total cost, not whether it used the fewest tokens.
Primary sources
Figures and product details above are dated and attributed; they are not current guarantees.
- OpenAI: LegalOn primary account ↗The attributed workflow and reported result. Proposed pilots and decisions below are original Roseram analysis.
The decision matrix
Original evaluation criteria for this workflow. These are questions to test, not results measured by Roseram.
| Decision | Evidence to collect | Stop or revise when |
|---|---|---|
| Routing | Task complexity and acceptance defined | Cheaper output needs disproportionate repair |
| Budget | Retries and escalation recorded | Limits silently stop a needed investigation |
| Outcome | Accepted changes and regressions measured | Generated diffs are counted as delivery |
YOUR EVALUATION / NOT A REPORTED RESULT
A small pilot you can inspect
Use these steps as a planning checklist. Checking a box records your progress in this view; it does not run or validate an experiment.
Questions this case answers
Is the reported cost reduction universal?
No. The account uses scoped estimates and measures; verify your own workload and billing.
Why not optimize only token usage?
Tokens do not capture retries, reviewer time, correctness or regression cost.
Tools & capabilities referenced
Operators studying this story often evaluate adjacent AI products. Explore Roseram guides where available:
PRACTICAL NEXT STEPS
Take the evidence into your own project.
Calculate the whole project cost
Compare subscriptions, usage, launch work and renewals.
Test an existing-project workflow
Check import, source loading, preview and publication.
Verify the output you buy
Require evidence tied to a real customer task.
Wiki brief
LegalOn is examined here for coding-agent cost controls and accepted output. The article answers “How can a software team control AI coding spend without lowering delivery quality?” using linked primary material and original evaluation design. It is not a customer testimonial for Roseram or an endorsement by LegalOn.
Operator takeaways
- Define accepted work.
- Count failed attempts.
- Keep escalation available.
Related case studies
Plan → build → review with routed models and transparent usage.