I had to hit the brakes on our AI outbound

I'm Jeremy Hurst, VP of Autonomous GTM at Swan. I left a $450K VP Sales role to take a job with a made-up title. I did it because I got a glimpse of the 100x seller - it stopped being a thought experiment and became the actual playbook. This is a field journal for GTM and growth leaders who can feel the model changing under their feet and prefer changing with it to being left behind.

AI agents are sending three quarters of our outbound messages. And they're getting 30%+ reply rates. But I still had to hit the brakes last week.

first, some context. we run a 4-tier lead scoring system. diamond, gold, silver, bronze. bronze & silver have been fully autonomous for a while. then gold joined the party last week.

when i made the change, i also started watching more closely. and i'm glad i did. because what i found was a complete mess.

3 separate documents trying to govern the same thing: how the agent writes & sends outreach.

scenario coverage was a tangle. one page handled leads from stalled trials. a separate page handled post-trial sequences. neither referenced the other.

multi-threading instructions - how to loop in senior execs - existed in 2 places at once. a dedicated page, and then again, embedded inside individual scenario pages. same topic. different guidance.

the whole thing was full of circular references. pages pointing to other pages that pointed back. instructions that contradicted each other depending on which document the agent loaded first.

a complete mess. and it was causing serious messaging drift. so i rebuilt it from scratch.

one parent skill: shared rules that apply every time. who we target, how sequences are structured, which sender handles which scenario, what needs approval.

then scenario sub-pages under the parent: one per situation. what's the context, what are we trying to accomplish. no voice instructions here.

finally, sender voice subpages: one per person. how they actually write. no scenario logic here.

the important thing: scenarios & voice are completely separate. the agent loads one of each and composes them at runtime. they never conflict because they were never in the same place.

and it turns out, this answers a question i keep getting from sales leaders.

how much of your outbound strategy should be centralized? and how much should you leave up to the reps? senior sellers have their own style, their own relationships. you don't want to strip that out. but you also have data on what works and you want to double down on that.

most people frame this as a tradeoff. it's not.

the parent skill & scenario pages are your centralized layer. overall approach, sequence structure, approval logic, what's working. enforced every time, for every sender, in every situation.

the voice subpages are your autonomy layer. each rep owns how they write. their personality, their rhythm, their style. none of that is touched by the centralized layer.

adding a new rep doesn't touch the scenarios. updating what's working doesn't touch anyone's voice.

you centralize the what. you decentralize the how.

getswan.com

THE SIGNAL

The centralization debate has an answer. Centralize the what, decentralize the how. The architecture enforces the rest.

THE PLAY

The Architecture Rebuild

• Pull every document your agent uses to govern outbound. List them.
• Find every topic covered in more than one place. That's your conflict map.
• Build one parent skill. Targeting rules, sequence structure, approval logic, what's working. Everything that applies every time goes here.
• Create one scenario sub-page per situation. Context and goal only. No voice instructions.
• Create one voice sub-page per sender. How they actually write. No scenario logic.
• Stress-test the architecture: add a new rep. If you have to touch the parent or any scenario page to do it, the structure isn't clean yet.

SIT WITH THIS

If you added a new rep tomorrow, how many documents would you have to update?

-Jeremy

[Read more.

](https://www.linkedin.com/posts/jeremyhurstprofile_ai-agents-are-sending-three-quarters-of-our-activity-7472625007282569216-zQOB?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAYm8JkB3GSdkfgU5-3_tdW6c48fsSkXvCs)