The five best ways to do this job with AI this month, tested by Redrob on real work from the profession. Pick one to use it in the chat.
Top 5 for a lawyer drafting contracts
#1 for draft contractsBest pick
Drafts from the term sheet, your templates and past deals in the folder, and keeps defined terms consistent.
What it was asked
Draft a master services agreement from the attached term sheet (Seorin Construction, KRW 2.1bn over 24 months) on our MSA template. Keep our standard liability and IP positions and flag anything the term sheet leaves open.
What it wroteSample output, illustrative
How this score was calculated
| Dimension | Measured | Score | Weight | Points |
|---|---|---|---|---|
| Quality | rubric 9.0 / 10 (experts 9.2, judges 8.8) | 90.0 | × 0.55 | 49.5 |
| Reliability | 96 of 100 runs passed | 96.0 | × 0.25 | 24.0 |
| Speed | median 5.0 min a task | 89.4 | × 0.05 | 4.5 |
| Cost | $16 a month | 63.8 | × 0.15 | 9.6 |
| Score | 95% interval ±1.2 | 87.5 |
Score = 0.55 Quality + 0.25 Reliability + 0.05 Speed + 0.15 Cost. Speed = 100 x sqrt(fastest median / this median). Cost = 100 x sqrt(cheapest month / this month).
Graders’ note Used the house template and kept every standard position; defined terms consistent throughout.
Draft method. Each monthly edition runs every combination of model, harness and effort Redrob can reach on 20 prompts per task, five times each. The prompts are written with people who do the work, from the task’s own profession.
Quality: two professionals grade every output blind against the task’s rubric, and three model judges from makers other than the one being graded grade it too. Quality is 60% the professionals and 40% the judges, shown with how far the judges agreed (Cohen’s kappa).
Reliability is the share of the 100 runs that reached 7 out of 10 on the rubric. Speed and cost are scored against the best in the task on a square root, so a cheap or fast model cannot win on that alone. Weights differ by task and are shown with every score.
The interval is a 95% bootstrap over prompts and runs; where two intervals overlap, the order between them is not settled. Prices are a typical month of this task at each maker’s published API price, before discounts, in the currency of the page’s language.
In this edition the figures and the output excerpts are sample content written for the design system, not real runs. Check every citation and clause before you rely on it.