Stop Writing SOPs Nobody Reads. Start Writing Playbooks Agents Run.

A practical operator guide to the 12-part Agent Playbook — with a GovCon capture workflow, a before/after conversion, and a Monday move you can execute this week.

I have read a thousand SOPs.

I have followed maybe three.

That is not a knock on the people who wrote them. It is an honest audit of how standard operating procedures actually live in most organizations — buried in shared drives, frozen at the moment of their creation, and consulted only when an auditor is already in the building. SOPs were built for a human-only operating model. Long checklists. Passive voice. Step 4 assumes the reader already knows what Step 3 actually meant.

That model is ending. And if you run any kind of services operation — especially in GovCon, where repeatable-but-rigorous work is the entire business — this shift is worth your full attention right now.

An SOP documents a workflow. An Agent Playbook executes it.

That is the whole distinction, and it matters more than it looks.

Here is the same task, written both ways.

SOP version (old): “Review the RFP and identify compliance requirements.”

Agent Playbook version (new): “Analyze the RFP. Extract all mandatory requirements. Classify each by section. Map each requirement to a response owner. Flag ambiguities for human review. Generate a compliance matrix. Queue it for the capture lead before end of day.”

One is guidance. The other is executable operations. One requires a human to remember, interpret, and apply it under deadline pressure. The other runs.

That is not a small upgrade. That is a category change.

The first principle underneath all of this

Any repeatable knowledge workflow can be converted from documentation into instruction — and from instruction into agentic execution.

Read that again slowly, because it is the operating thesis for the next three years of services work.

This does not mean automate everything. It means examine everything. Ask six questions about every workflow you own:

1. Is the work repetitive?
2. Where is human judgment genuinely required?
3. Where is evidence needed to defend a decision?
4. Where do errors happen most?
5. Where does cycle time cost you bids or revenue?
6. Where does quality vary based on who performs the task?

Wherever you answered yes to more than two — that is a candidate for an Agent Playbook. The goal is not to replace judgment. It is to compress the distance to judgment.

Why SOPs actually break down

Before you can convert them, you have to understand why they fail in the first place.

SOPs are too long. They are outdated within six months of being written. They live in folders no one opens. They assume context the reader doesn’t have. They describe ideal behavior, not actual behavior. And they depend on humans to remember every step — while also managing a deadline, an inbox, and a client call.

SOPs are static. Work is dynamic.

An Agent Playbook solves this structurally. It encodes the workflow. It retrieves context. It executes steps. It produces artifacts. It routes exceptions. And it preserves a complete audit trail of what happened, when, and why.

That last point matters enormously in regulated environments. Governance is not a feature you bolt on — it is built into the playbook architecture itself.

The 12-Part Agent Playbook — A Usable Template

This is not prompt writing. This is operational architecture. Every Agent Playbook you build should define all twelve of these components. I use this as a literal checklist before I ship any new playbook into production.

• [ ] Mission — What is the purpose of this workflow in one sentence?
• [ ] Agent role — What persona, function, and authority does the agent carry?
• [ ] Inputs required — What documents, data, or signals does the agent need to begin?
• [ ] Trusted sources — What repositories, databases, or feeds is the agent authorized to pull from?
• [ ] Tools available — What functions can the agent call? (web search, doc parser, CRM write, calendar, etc.)
• [ ] Steps to execute — The ordered sequence of actions, written as commands, not descriptions.
• [ ] Output format — Exactly what artifact does the agent produce? (memo, matrix, scored list, draft section)
• [ ] Quality standard — What does “good enough to hand to a human” look like? Define it explicitly.
• [ ] Escalation rules — What triggers a hand-off? What conditions require a human before the agent continues?
• [ ] Human approval points — Which outputs require a sign-off before they become inputs to the next step?
• [ ] Audit requirements — What does the agent log? Where? What is the retention rule?
• [ ] Feedback loop — How does the Agent Operator improve the playbook over time?

Twelve components. Most SOPs cover maybe four of them implicitly — and none of them explicitly enough to hand to an AI agent. This template is the bridge.

GovCon capture: a before/after you can copy

This is where it gets concrete. If you are in government contracting, you are running capture operations right now against a September 30 government fiscal year close. Bid/no-bid speed is not a nice-to-have — it is a competitive moat. Slow qual is lost task orders.

The old SOP said: “Monitor relevant contract vehicles. Review new opportunities. Assess fit against capabilities. Prepare a bid/no-bid recommendation for leadership.”

Vague, human-dependent, variable in quality based on who ran it that week.

An Agent Playbook executes this instead:

1. Monitor target IDIQ vehicles and GSA schedules on a defined cadence
2. Pull each new opportunity: agency, NAICS code, scope summary, ceiling value, period of performance, incumbent, set-aside status
3. Compare requirements against capability library and past performance database
4. Score strategic fit on a 1–10 rubric (you define the rubric once; the agent applies it consistently)
5. Identify likely competitors from SAM.gov award history and known teaming relationships
6. Flag missing information for the capture lead
7. Draft a bid/no-bid recommendation memo with supporting rationale
8. Queue the memo for executive review before 24-hour response window closes

The human still decides. Full stop — that authority does not move.

But the human no longer starts from a blank page at 11pm. The agent compresses what used to take a team 8–12 hours into a standing workflow that runs on schedule. These are the kinds of operations — opportunity monitoring, agency research, capture intelligence, compliance review, proposal support, strategic account planning — that turn growth operations from manual heroics into repeatable intelligence systems. The label matters less than the architecture.

This is also your training system

Here is an implication most operators miss: when you build the playbook right, you get onboarding for free.

New hires see exactly how the workflow runs, what sources are used, what good outputs look like, where humans intervene, and how decisions are made. In the old model that took six to twelve months of shadowing. In the new model the Agent Playbook is the tribal knowledge — explicit, executable, inspectable. Institutional memory lives in the playbook, not in the person who is about to resign.

The governance advantage nobody talks about

A well-designed playbook makes your workflow auditable by default.

In a regulated, compliance-heavy environment — which is GovCon, healthcare services, financial services, legal ops — “how does this process work?” should have one answer. Not twelve slightly different versions depending on who you ask.

The playbook is the answer. Leaders can inspect it, improve it, and certify it. When an auditor asks what the agent was authorized to do, what sources it used, what outputs it created, and what human approved the final decision — the playbook is your evidence trail.

That is governance by design, not governance by hope.

The Monday move

Pick one SOP this week. Not your best one — the one everyone quietly ignores. The one that exists because someone wrote it three years ago and now it mostly lives in a folder nobody opens.

Run it against the 12-part checklist above. You will find, I promise, that it covers maybe four of the twelve. That gap is your work. Close it. Write the playbook version. Test it on one real workflow.

That is the conversion. That is how the operating doctrine for intelligent work gets built — one playbook at a time, by operators who take ownership of the systems they run.

The firms that build this architecture in the next two to three quarters own the institutional memory advantage. The ones still running on static SOPs will be coaching their agents with documents their agents cannot follow.

I run this playbook discipline at Trust Income Equity. If you are building an AI-native services or GovCon operation and want to compare notes, I will make time for that conversation.

— Corey Solivan