The Great Implementation
Convert documented human playbooks into reusable AI-assisted workflows.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 95%
The Great Implementation is the operational phase in which companies translate existing playbooks into AI-assisted systems. A team starts with a documented workflow, isolates a bounded task that the model can perform, and proves its reliability. It then surrounds that model call with the integrations, account connections, templates, and human checkpoints needed for regular use. Instead of expecting an AI to occupy an entire job, the organization packages repeatable capabilities so existing workers can clone and run them. This implementation layer determines whether raw model capability produces actual productivity. As templates spread, inexperienced workers gain leverage faster, small companies can postpone specialist hiring, and jobs change through the gradual absorption of their documented constituent tasks.
Origin
Nathan Labenz called the transition the great implementation while describing AI-assisted executive-assistant and recruiting playbooks on Marketing Against The Grain.
Core principles
- 01Documented workflows are strong automation candidates.
- 02The core AI task may be easier than the surrounding implementation.
- 03Reusable templates accelerate adoption across users.
- 04AI can augment workers before it eliminates whole roles.
How to run it
- 1
Choose a Proven Playbook
Select a workflow people already repeat and understand, preferably one with written instructions and examples.
Pro tip Start with a process that users already want to delegate.
- 2
Extract the Core Task
Identify the model-sized activity inside the playbook, such as evaluating candidate fit or drafting outreach.
Pro tip Separate reasoning from account access, notifications, and approvals.
Watch out Do not send an entire end-to-end role to the model as one prompt.
- 3
Prove Reliability
Test the core task against representative examples and edge cases before automating it.
Pro tip Use the same evaluation criteria currently applied to human work.
- 4
Build the Workflow Wrapper
Connect the model to the required tools, data sources, and destinations through an automation platform or application.
Pro tip Minimize the number of manual handoffs while preserving review where stakes are high.
Watch out A good prompt alone is not a production workflow.
- 5
Template and Distribute
Package the automation so workers can clone it, connect authorized accounts, and adapt a small number of settings.
Pro tip Provide safe defaults and clear failure handling.
- 6
Improve from Use
Monitor output quality, exceptions, and time savings, then revise the playbook and automation together.
Pro tip Treat operational failures as feedback about both the workflow and the model.
In the wild
An executive-assistant company takes its documented passive-candidate sourcing methodology and isolates candidate evaluation and personalized outreach. Those tasks are implemented as template automations that assistants can clone and connect to client accounts, while assistants oversee targeting and exceptions.
→ New assistants execute sophisticated recruiting workflows faster and with more consistent quality.
Common mistakes
Confusing a Prompt with a System
A successful model response does not provide integrations, monitoring, permissions, or exception handling.
Automating an Unstable Playbook
If humans cannot explain the process consistently, packaging it will reproduce confusion at greater speed.
Removing Review Too Early
Consequential hiring or outreach decisions still need checkpoints until reliability is demonstrated in practice.
Is it for you?
Best for
Companies with mature playbooks, recurring administrative processes, and accessible automation infrastructure.
Not ideal for
Undocumented work that changes every time or requires extensive tacit organizational knowledge.
From the transcript
“Those sorts of highly documented, highly defined tasks really lend themselves well to AI completion.”
“And so I think we're, you know, I called this the great implementation in that thread.”
“I think we're going to be bundling those up into zaps and kind of making them, you know, template zaps that EAs can go and…”
From the episode
GPT-4 Beta User Reveals What Jobs It Will Destroy In 2023 with Nathan Labenz (#103)
Nathan Labenz