Field Notes
July 23, 20266 min read

Three Kinds of Agent Workflows - and How to Pick Your First

Justin Henriksen
Justin Henriksen

Founder & CEO, GetLatest AI

A contractor I know spent $400 on Google ads last month. Got 11 form submissions. Followed up on 3 - the ones she happened to see before her phone died at a job site. The other 8 sat in her inbox until they weren't worth calling anymore.

The ads worked. The follow-up didn't.

And her situation isn't unique - it's the same shape I see everywhere. Form fills. Someone checks the inbox when they get a chance. Pastes the company name into Google. Pulls up the CRM to see if they're already there. Tries to draft a response that sounds personal.

That sequence happens inconsistently, at different speeds depending on who's available. And the variability costs real money. Leads that get a response in 5 minutes convert at dramatically higher rates than leads that get a response in 5 hours. Most small businesses are operating in hours.

The shape of this problem is identical across industries - professional services, SaaS, home services. Trigger, decision, handoff, output. The specific systems differ. The pattern doesn't.

That's what a playbook is: the documented pattern you adapt to your situation, not a script you run unchanged.


The three categories

Agent playbooks fall into three groups based on where in the growth process they operate.

Marketing playbooks are mostly observational. The agent watches something on your behalf, synthesizes what it finds, and surfaces the result for a human decision. Monitoring what appears about your business in AI-generated search results. Tracking ad performance against the signals that actually predict conversion. Managing review presence across platforms. The time and consistency savings are real even when the agent has no authority to act - you're getting synthesis that otherwise wouldn't happen, or would happen on a slow manual schedule.

Sales playbooks operate closer to customer conversations, which is why the approval architecture is tighter. These cover: responding to inbound leads within minutes instead of hours, qualifying and routing them consistently, preparing for meetings with context gathered across your systems, following through on what meetings produce, monitoring pipeline health, reactivating deals that went quiet, building proposals with the right context assembled before the call.

Connected growth playbooks treat marketing and sales as one system. A prospect attends a webinar - the agent surfaces that signal to the sales rep with enough context to decide whether to reach out. A rep hears the same objection 12 times in one month - the agent surfaces that pattern so marketing knows what to address in content. Deal-stage and close-rate data becomes visible to the people deciding where to invest marketing budget.

That third category matters because the biggest inefficiency in most growing businesses lives at the handoff between marketing and sales. Information marketing collected never reaches sales. Patterns sales hears every week never inform what marketing creates. Connected growth playbooks address that gap directly.


What every playbook is honest about

A playbook isn't a demo. It's a starting point that tells you what the workflow looks like when it's working, what tends to go wrong, and what a human still needs to own.

Five things worth examining closely in any playbook before you build.

The trigger. What specific, observable event starts the workflow? A trigger that's too vague - "when something interesting happens" - can't be wired up. A trigger that's specific - "a new contact form submission arrives" - can be. If your version of the workflow doesn't have a defined trigger, it isn't automatable yet.

What the agent does vs. what the human decides. This is the most important section, and the one most often misread. People read a playbook and imagine full automation. What playbooks actually describe is a division of labor: the agent handles observation, synthesis, and first drafts; the human handles judgment, approval, and consequential action. The agent reads the lead and proposes a qualification score. The human confirms or corrects. The agent drafts the follow-up. The human sends it. That division isn't a limitation waiting to be removed - it's the design.

The approval points. Where must a human confirm before something consequential happens? A summary for internal review doesn't need approval. An email to a prospect does. The playbook tells you where the gates belong.

How you know it's working. Not "feels faster." Specific rates: first meaningful response time, qualification acceptance rate, manual effort per opportunity. You need a baseline before the agent runs. Without it, you can't know whether the agent improved anything.

What breaks it. The form submission arrives with no company name. The lead is already in the CRM under a slightly different email. The assigned rep is on vacation and nobody set a fallback. Knowing the failure modes before you build is part of knowing what you're building.


Where to start

Two playbooks have consistently faster measurable impact than the rest.

Inbound lead response has the fastest feedback loop. A lead comes in, something happens or doesn't, and you can measure the result within hours or days. If your current process is slow, inconsistent, or dependent on whoever checks the inbox first - an agent that produces a scored summary and a draft response within minutes of submission creates visible improvement almost immediately.

The first test doesn't even require connecting anything to production. Run the workflow manually, use AI to handle the scoring and drafting steps, and measure how long it takes and what the response quality looks like. That's your baseline. Then you have something to beat.

Stalled opportunity reactivation has the highest immediate revenue potential for businesses with existing pipelines. If you have deals in your CRM that went quiet, the work of reconstructing context and identifying the right next action is exactly where agents add value. Most businesses that run this workflow honestly find that a meaningful portion of their "stalled" pipeline should just be closed as lost. That clarity is itself valuable - those deals are taking up mental space that could be on something real.

How to choose between them: where is the bigger, more immediate problem? Healthy pipeline but inconsistent lead handling? Start with inbound response. Pipeline full of deals nobody is working? Start with reactivation.

Start with one. Get comfortable with what the agent produces. Measure it. Move to the next one when the first is running reliably.


One thing to do this week

Look at the triggers for two playbooks: inbound lead response and stalled opportunity reactivation. For each one, ask: does that trigger apply to my business as described, and if not, what's my version of it?

If you can define your version of the trigger, you're closer to being ready to build than you think.


Want me to look at yours?

Bring one workflow to a 30-minute call with me. I'll tell you exactly where an agent would fit in your business, what it would take to build, and whether it's even worth it for you. No pitch - just an honest read from someone who does this for a living.

Book a mapping call

Justin Henriksen

Justin Henriksen

Founder & CEO, GetLatest AI

Justin is the founder of GetLatest AI. 25 years building and leading technology, from Principal SWE to CEO. He writes about AI agent architecture, production systems, and what actually works.

Want your AI team built and run for you?

Book a free call30 min · no commitment