Resources / Take something useful
Start with the AI readiness checklist if you’re working out where to begin. Use the briefs and scorecards to define a first project and check whether it helps. The scripts and playbooks offer more detail for builders. All are available to use independently.
01 / Ready to use
Practice the same account-update task with complete and incomplete fictional sources, reusable instructions, and a human review checklist.
Download .mdDefine deliverables, source access, testing, handoff, and ongoing costs before agreeing to an AI project.
Download .mdEvaluate evidence, responsibilities, review, and ownership before choosing a consultant.
Download .mdCompare a clearer process, a built-in feature, and a custom workflow using one representative task.
Download .mdPlan a practical session around your team’s questions, a permitted sample task, and an output people can review.
Download .mdReview today’s AI use, choose a relevant first task, and agree on training, information access, and ownership.
Download .mdDefine one business workflow: its inputs, permissions, review steps, and a contained first test.
Download .mdCompare the complete task, record corrections and cost, and decide whether the pilot is worth continuing.
Download .mdMap the work, define success, and scope a contained first AI experiment.
Download .mdGive an agent a clear outcome, the right context, and boundaries it can follow.
Download .mdTurn a recurring task into a prompt with defined inputs and a review step.
Download .mdTHE AI AGENT PLAYBOOK / FREE & OPEN SOURCE
Developer and founder guides, specialist agent templates, hooks, and a starting constitution for your AI development team.
Explore the playbook02 / A small working script
Track active preparation and initial work, review, corrections, and pass/fail results in a local CSV. Keep unattended waiting in elapsed turnaround, recorded separately on the pilot scorecard.
Download the script Download the usage guide Record elapsed time with the scorecardpython3 workflow-run-log.py log \
--csv ./experiments.csv \
--workflow "Meeting follow-up" \
--run-minutes 2.5 \
--review-minutes 4 \
--correction-minutes 1 \
--result pass
python3 workflow-run-log.py summary \
--csv ./experiments.csvPreparation + active run: 2.5 min
Review: 4 min + Correction: 1 min
Total active effort: 7.5 min
03 / Keep these close
Name three things a useful result must get right. For meeting follow-ups, try decisions, action owners, and due dates. Use those checks on every test so a polished draft cannot hide a missing detail.
Try it with a resourceTell the model what to write when a required detail is absent: ‘not provided,’ ‘needs confirmation,’ or a clear question. A visible gap is easier to resolve than a plausible guess inside a confident answer.
Try it with a resourceReading, drafting, and sending are three different permissions. State which actions are allowed in the handoff and name the point where a person reviews the result. The agent gets useful freedom inside a clear boundary.
Try it with a resourceTrack checking and corrections alongside the initial run. Record failed attempts too. The useful comparison is the effort required to finish the whole task, using the same timing method for both your current process and the experiment.
Try it with a resource04 / Open-source work
Plain-language setup, troubleshooting, context/token management and scheduled workflow guides for nontechnical teams.
Code examplesPublic Python examples for prototyping Reddit API calls.
ScriptsPublic Python functions for editing and compressing web video.
SystemPublic repository describes an agent operating system for agency delivery with queues, review and human-controlled release.
CLIPublic repository for auditing, improving and simulating ecommerce product detail pages.