The Ramp and Factory examples are compelling - but they're companies that built the culture and structure alongside the tooling.
The harder version of this question is what happens in organisations that didn't start AI-native. Same L0-L3 ladder, but L3 systems builders are trying to encode governance into workflows that weren't designed for it.
"Automate your job" is clean advice when you own your job. It's harder when three teams share the process and nobody owns the outcome.
This was a wonderful article. Thank you for taking the time to thoughtfully putting this together. A lot of articles are talking about the "what" where they were just saying yeah we use AIs. Lol, who's not. We all also get the "why". But this article discusses the "how". We need more of the "how" articles just like this one. Looking forward to your next one!
I called this "Vibe Management" a couple of weeks ago. If you treat your agent like an employee, things go a lot smoother. The key is providing context. I am working on an agentic model where high level business knowledge such as brand identify, strategy, target customers, etc is stored in a shared data store. Each agent can then be asked or programmed to check if their actions align with the company's goals. If not, that issue is flagged and raised to the appropriate stakeholder.
This is the operational shift a lot of teams still underestimate. Once agents enter the workflow, onboarding stops being an HR metaphor and becomes a systems problem: permissions, context quality, escalation paths, and review cadence. The orgs that win will be the ones that make those handoffs boring and reliable.
Strong framing. The companies getting the most from agents seem to treat onboarding as systems design, not prompting theater. Clear task boundaries, shared context, escalation rules, and post-task review loops matter more than “which model won this week.” That operational layer is where the compounding happens.
I love this. That said, how do you approach this inside a company that controls what you can use? I've spent the past year figuring out how to get better at AI with the tools I'm actually allowed to have. I'm writing about it at Constrained Intelligence, but I would love to get your take!
Ultimately, all of what has been shared is only possible when Leadership is fully aligned with encouraging and providing the resources for adopting AI in your daily workflows. It’s beyond just “telling” them. One way is to surface this article, quantify some opportunities and tie it to some tasks/problems that can be easily solved. It’s a trade off; will they rather spend 6 months shipping a feature when implementing processes like this can shorten feedback cycles and time to market?
Maybe also as a follow up, how does this framework apply to people on the business side (finance, strategy, HR, legal) at companies that didn’t start AI-native, is the L0–L3 framework still the right mental model? Or does “automate your job” look fundamentally different when you don’t own a codebase and your tools are whatever IT approved?
Asking because I work in exactly that environment and would love to know if you’ve seen this play out well anywhere outside of product/eng functions.
This makes a compelling case for “agent-first,” but it also begs a harder question: if everyone’s job is to automate their job, where does institutional judgment, apprenticeship, and genuine domain mastery get built over time? At what point does optimizing for token consumption and AI proficiency risk turning high-agency operators into prompt routers rather than strategic thinkers?
Such great and valuable actionable information and guidance! Thank you! :) More like this please
The Ramp and Factory examples are compelling - but they're companies that built the culture and structure alongside the tooling.
The harder version of this question is what happens in organisations that didn't start AI-native. Same L0-L3 ladder, but L3 systems builders are trying to encode governance into workflows that weren't designed for it.
"Automate your job" is clean advice when you own your job. It's harder when three teams share the process and nobody owns the outcome.
This was a wonderful article. Thank you for taking the time to thoughtfully putting this together. A lot of articles are talking about the "what" where they were just saying yeah we use AIs. Lol, who's not. We all also get the "why". But this article discusses the "how". We need more of the "how" articles just like this one. Looking forward to your next one!
I called this "Vibe Management" a couple of weeks ago. If you treat your agent like an employee, things go a lot smoother. The key is providing context. I am working on an agentic model where high level business knowledge such as brand identify, strategy, target customers, etc is stored in a shared data store. Each agent can then be asked or programmed to check if their actions align with the company's goals. If not, that issue is flagged and raised to the appropriate stakeholder.
This is the operational shift a lot of teams still underestimate. Once agents enter the workflow, onboarding stops being an HR metaphor and becomes a systems problem: permissions, context quality, escalation paths, and review cadence. The orgs that win will be the ones that make those handoffs boring and reliable.
Strong framing. The companies getting the most from agents seem to treat onboarding as systems design, not prompting theater. Clear task boundaries, shared context, escalation rules, and post-task review loops matter more than “which model won this week.” That operational layer is where the compounding happens.
I love this. That said, how do you approach this inside a company that controls what you can use? I've spent the past year figuring out how to get better at AI with the tools I'm actually allowed to have. I'm writing about it at Constrained Intelligence, but I would love to get your take!
Ultimately, all of what has been shared is only possible when Leadership is fully aligned with encouraging and providing the resources for adopting AI in your daily workflows. It’s beyond just “telling” them. One way is to surface this article, quantify some opportunities and tie it to some tasks/problems that can be easily solved. It’s a trade off; will they rather spend 6 months shipping a feature when implementing processes like this can shorten feedback cycles and time to market?
Maybe also as a follow up, how does this framework apply to people on the business side (finance, strategy, HR, legal) at companies that didn’t start AI-native, is the L0–L3 framework still the right mental model? Or does “automate your job” look fundamentally different when you don’t own a codebase and your tools are whatever IT approved?
Asking because I work in exactly that environment and would love to know if you’ve seen this play out well anywhere outside of product/eng functions.
This makes a compelling case for “agent-first,” but it also begs a harder question: if everyone’s job is to automate their job, where does institutional judgment, apprenticeship, and genuine domain mastery get built over time? At what point does optimizing for token consumption and AI proficiency risk turning high-agency operators into prompt routers rather than strategic thinkers?
So much good info in here! Thanks for sharing how organizations are adapting!