Playbooks
You create the playbook once (based on how the lawyers in the firm already work or want the AI to behave and respond). Then the AI uses it every time, so the results stay consistent and closer to what the firm wants.
- Grok – Skills
- ChatGPT – Custom GPTs / Projects / Custom Instructions
- Claude – Projects
- Gemini – Gems
How?
- Do the task/document/contract/review once the way you want it done.
- When you’re happy with the result, say:
“Save this process as a Skill called Contract Review Skill.” - Then when you want to use the playbook or skill:
“Use my Contract Review Skill on this agreement” - You can edit the skills later as you wish under Plugins Skills
Why
Stops Prompt Repetition: Instead of re-explaining formatting rules or risk posture in every new chat, a Playbook locks in the firm’s baseline instructions.
Enforces Uniform Style: Standardises clause numbering, defined terms, and legal registers across the entire practice.
Reduces Review Bottlenecks: Generates cleaner first drafts that require significantly fewer structural redlines during partner review.
Things to consider when creating playbooks:
Document Review
- Extraction Rules: Which sections to target, such as finding payment terms, liability caps, or termination clauses.
- Risk Ranking Levels: Define what makes a clause high, medium, or low risk (e.g. unlimited liability is automatically high risk).
- Standard Fallbacks: Provide the exact replacement text or acceptable alternative range the AI should suggest when a clause deviates from firm standards.
- Formatting Guardrails: Instruct the AI to insert comments explaining why a redline is made so opposing counsel understands the rationale.
Core Elements for Contract Creation
- Template Source Control: Restrict the AI to specific, approved firm templates or a designated internal knowledge base rather than letting it invent structure.
- Intake Variable Mapping: Create a checklist of required inputs (e.g., party names, effective dates, governing jurisdiction, payment amounts) that the AI must verify before drafting.
- Completeness Checks: Program the playbook to scan the final draft for missing definitions, broken cross-references, or unassigned obligations.
- Tone and Style Rules: Specify preferred language standards, such as plain-language rules or specific definitions favoured by the firm.
Example:
Name: Master Agreement Drafter
Description: Custom Playbook for generating baseline commercial agreements following firm risk policies.
Instructions:
- Role: Senior legal draftsperson specialising in commercial contracts.
- Output Format: Clean, professional legal prose divided into numbered sections. Use markdown headings. Every variable data point must use square brackets like [PARTY A].
- Drafting Style: Clear, unambiguous, and concise. Avoid archaic legalese (e.g., “heretofore,” “witnesseth”) where plain legal English suffices.
- Risk Allocation: Default to protecting our client’s position regarding intellectual property ownership, indemnification caps, and termination for convenience rights.
- Safety Check: At the end of every draft, output a brief “Attorney Risk Notes” section highlighting any high-liability clauses that require human strategic review.
~
Create a playbook in conjunction with the other staff involved, the more input the better.
Revise the playbook every month.
Schedulers
In the past if you wanted an LLM to run every morning, you had to write custom code or use third-party automation tools like Zapier or Make to trigger an API call.
Grok – called Automations. – July 2026 (revised mainstream)
Stumbled upon it when asking it to teach me a Portuguese word and then asked it to do it each day. Now get a new word each day, with pronunciation, audio example, spelling and usage in a sentence. Emailed to me.
Other LLM Automation:
- ChatGPT: Tasks – June 2026
- Claude (via Claude Cowork / Scheduled Tasks): Paid plans (Pro, Max, Team, Enterprise) – March 2026
- Microsoft Copilot: Includes a direct prompt scheduling option in Copilot Chat where you hover over a prompt, click a clock icon – Dec 2025
- Gemini is called Scheduled Actions – June 2025
How?
After you have got the result from a prompt chat, ask it to do this everyday day, and at what time, it creates it in Automator.
Examples:
- Daily briefing – “Every weekday at 8am give me a short briefing: top news, weather in Durban, and any calendar events happening.”
- Share Price monitor – “Check Bitcoin and Ethereum prices every hour between 9am and 5pm and alert me if either moves more than 3%”
- Email Triage – “When I receive an email from @important-client.com, summarize it and draft a polite reply.”
- Once off reminder – “Remind me at 2pm tomorrow to run the SAB report.”
- Weekly review – “Every Sunday at 7PM, summarise what I achieved in the week based on my Linear Issues, emails, meetings and notes.”
- Client Follow-up nudges – “List any open client matters where I haven’t had contact in more than 10 days. Draft short, professional check-in messages for each.”
Other AI News:
Powerful AI on your own machine. Encrypted. Open source. Privacy-first architecture.
Free, then paid for Online Frontier models. Windows PC Spec says any modern Windows system – obviously, the bigger models would need more RAM.
