How Luro's CPO Built a Company-Wide AI Operating System That Turns Every Employee Into a Builder
Luro has constructed what Jess calls a "company OS" in GitHub: folders for every business function (CS, design, engineering, finance, legal, marketing) each containing skill files for every activity that function performs. Those files are uploaded into Claude's organization settings so employees can invoke the right skill without leaving their existing workflow. The practical outputs include a daily Slack briefing for all customer-facing staff, a Slack-native feature-request automation that eliminates back-and-forth and auto-creates triage tickets, and a mega go-to-market agent that routes requests to the correct sub-agent on demand.
The core organizational problem Jess is solving is a 1%-vs.-99% adoption gap: a small cohort of tinkerers runs far ahead while the majority of the company doesn't know what to use or when. Her answer is a four-level AI proficiency framework (chat → automate a workflow → build personal apps → ship shared apps), a function-by-function work ontology that distinguishes tasks to automate from tasks to keep human, and a deliberate decision to deliver everything through Slack and email rather than a separate interface—because meeting people where they already work is what drives actual adoption.
On the product-building side, Luro has moved to a "captain" model where the person whose skill set is most critical to an outcome owns a feature end-to-end, including QA. Non-engineering PMs and even a customer success manager have shipped front-end and back-end features to production using Devin, guided by an internal enablement playbook. A lightweight "ask Devin reviewers" Slack channel provides visibility and appropriate expert review without reinstating a waterfall process. Larger features requiring broad alignment still go through formal product strategy and architectural review, preventing the failure mode Jess calls "everyone running fast in different directions."
Key concerns raised: AI adoption remains lumpy across functions, especially in go-to-market roles. Scheduled automations proliferate into information overload if not consolidated. And many experienced PMs are responding to the transition with fear rather than curiosity—a bifurcation Jess sees as defining who thrives. Relationship-building (real on-sites, champion dinners, genuine hospitality) is explicitly called out as irreplaceable by agents.
Jess's bottom-line prescriptions: assign a dedicated AI ops person with a full-time charter rather than making AI "everyone's responsibility" (which means no one's); run quarterly hackathons to normalize building across all functions; color-code your team's work map to identify what should be AI-enabled immediately; and in hiring, ask candidates to screen-share their live AI usage to quickly gauge whether they are at Level 1 or above. The fundamentals of product management—knowing why and for whom you build, and what success looks like—have not changed; the speed and tooling around them have changed radically.
Jess, CPO at Luro, details how her team built a structured AI operating system—skill files, playbooks, and multi-agent workflows—to close the gap between the 1% of 'AI-pilled' power users and everyone else. She covers how non-engineers are now shipping production features via Devin, and why customer judgment, not speed, remains the irreplaceable PM skill.
Creating too many scheduled automations leads to information overload; consolidating all automations into one company OS delivered via Slack or email just-in-time is more effective than individual scheduled tasks
Delivering just-in-time playbooks and automations inside Slack and email, rather than requiring users to visit a separate interface, is the key to driving actual agent adoption
these are the parts that remain human-centric versus these are the parts that should be automated away.
The most important thing about the PM role that will never change is staying close to customers — knowing why and for whom you are building, what you are trying to solve, and what success looks like.
if you are in a PM job where everything we were just talking about feels really foreign and like 10 steps away from what you are, find a job like this with an AI PM CPO like JZ.
90 to 99% of the rest of the organization who isn't sure what to use when
Luro has built a company-wide operating system in GitHub with folders for every single function: customer success, data science, design, engineering, finance, legal, marketing, each containing skill files for every activity that function performs
Every single morning, every person on customer-facing teams receives a daily briefing showing their calendar, meetings, check-ins, and onboarding sessions via Slack
The ontology for every function maps all work to categories and tasks, and has been built out for marketing, sales, customer success, implementation, design, engineering, and product
The work map of a product manager is starting to look a lot more like an engineer
PM tasks like competitive market analysis, writing for stakeholder management, synthesizing feedback are starting to get automated, but in a lumpy way where one PM might be doing it well and another might not
Luro's customer success playbook is 50 pages covering implementation, onboarding, user onboarding, and different tracks for admin vs. timekeeper personas
Playbooks that used to take a long time to write can now be created very fast with Claude from a lot of sources
Luro uses Dust as its agent-building tool, starting in fall of last year, because it was easier to use a specialized agent builder like Glean or Dust at that time
The gap between specialized agent builders like Glean or Dust and building agents directly in Claude is shrinking quite rapidly
The fundamentals and the principles have never changed. In fact, they're even more important than ever before, but the tools and the way you operate, that's radically changed.
Devin started off about a year ago, 2 years ago, when we first started using this as almost like intern-level engineer. And today, I think it it's actually, you know, a decent software engineer. It's not a staff-level software engineer, but it does a lot of things.
Nick, who is a PM on Jess's team and self-identifies more on the design side than on the engineering side, shipped a front-end and back-end feature (temporary initiatives) end-to-end using Devin.
Jessica, a PM on the team who did not start her career in engineering or study computer science, built the empty state new user onboarding experience end-to-end.
Ashley, a customer success team member, used the PM-created Devon enablement guide to ship features to production.
The temporary initiatives feature, which included front-end and back-end changes and integration with PMSs and other systems, did not go through a product review.
The product builder who ships a feature is personally responsible for end-to-end testing of that feature, eliminating the waterfall handoff loop between PM, designer, and engineer.
Laurel held a company-wide hackathon at their offsite 3 months ago and considers running hackathons every quarter or every 6 weeks to set the expectation that everyone in the company, not just engineering, is a builder.
Competitive analysis tasks — including synthesizing competitive market intelligence, writing detailed briefs, doing research planning, doing reach out for research, and synthesizing research — should be fully automated via agents, with PMs only moderating the output.
Notion used Customer.io to personalize their onboarding and hit nearly 50% open rate, improved conversion by 6 to 7% with localized campaigns, and pushed open rates up another 20% through AB testing.
Every company is going to have to get to AI-enabled workflows because competitors are moving at 10 times the speed.
A playbook first draft can be written in sub a minute with AI, and a complete, business-accurate version takes hours, maybe days max, not weeks.
When you say something is everyone's responsibility, it's no one's responsibility.
Laurel has a dedicated AI operations team, started with one person named Sasha, whose demonstrated value led every other function to request their own dedicated AI ops person.
Laurel's CEO Ryan re-architected the entire product and company to be AI native after foreseeing what timekeeping would look like in a world of LLMs.
Jess currently has five PMs and four designers and sees no real reason to grow that number because adding more people adds more coordination costs.
The best PMs are getting more roles while others feel fear because one AI-pilled PM with strong judgment can now do what previously required many people, but very few PMs actually have that combination of traits.
Jess teaches AI leadership through Reforge and the curriculum changes literally by the month, with massive change between each 6-month cohort.
Former CPOs and VPs of Product are among the most excited builders because they realize that large team management was mostly coordination overhead.
the speed has changed dramatically, but what you're supposed to be doing at the heart of it, that has not changed.
we have hit a crazy milestone when we crossed 40,000 YouTube subscribers
We have also crossed 565,000 average views per listen per episode.
I started this podcast 2 years ago
You got these people who are these 1% AI users. They're tinkering with their workflows. They're highly AI pilled, and then you have the 90 to 99% of the rest of the organization who isn't sure what to use when.
AI adoption is consistent among PMs and engineers but not across all functions, particularly go-to-market functions
a lot of quote-unquote AI-native companies are just like, roadmaps are gone, planning is a gone, everything is gone. Um, and what I say is, well, if everyone's running in different directions, even if you're running incredibly fast, you're not really going to get anywhere.
Relationship building — real check-ins, true hospitality and delight, actual on-sites, taking a champion out to dinner — cannot be replaced by agents.
Many experienced PMs are feeling fear rather than excitement about the AI transition, and Jess sees a bifurcation between those scared their job is changing and those who have never been more excited.
Four levels of AI use: Level one is chat mode (ChatGPT, Claude); Level two is automating a workflow; Level three is building apps; Level four is building shared apps
Company OS skill files can be uploaded into Claude's organization settings so that employees can call the right skill at the right time without leaving their existing workflow
Three steps to build a company OS: Step 1 - start small with one tedious repeatable workflow; Step 2 - create a playbook and identify which parts to automate vs. keep human; Step 3 - build agents for each playbook step and wrap them in a mega-agent that routes requests
Laurel uses a 'captain' model where the captain of any initiative is the person whose skill set is most critical to the outcome — engineering captain for architectural changes, designer captain for interaction-heavy features, PM captain for content and customer-understanding-heavy features.
Laurel uses a two-track model: small features can be taken end-to-end by a single product builder with lightweight review (e.g., an 'ask Devin reviewers' Slack channel and PR review), while larger features requiring broad alignment go through a formal product strategy review and architectural review.
Laurel built a company-wide ontology mapping every function's work into buckets, distinguishing time that should increase (e.g., feature work with agents, QA) from time that should decrease (e.g., manual competitive synthesis, research planning), and tracks actual time allocation against those targets.
Scale the workflows of your top 1% performers to every single person on the team, and celebrate those wins to create a culture where AI enablement is expected.
AI ops is a new biz ops — find people who are insanely curious, tinkering with the latest technology, and relentless about finding efficiencies, and give them a dedicated full-time charter.
Four AI proficiency levels: Level 1 = chat/search mode with ChatGPT or Claude; Level 2 = automating a workflow; Level 3 = building personal apps; Level 4 = shipping shared apps or shipping to customers.
Do you have the right culture? Do you have the right team? Do you have the right space for people to even build? Do you have the right operating system? Do you have the right knowledge of what people are doing day-to-day?
Luro created a Slack automation that automatically collects feature request information (frequency asked, Gong recording, customer impact, account manager judgment) eliminating back-and-forth, then auto-assigns triage and creates a tracking ticket
Luro built a mega go-to-market agent that can be called by the sales or success team at any point and routes the request to whichever sub-agent is most useful
Laurel created an 'ask Devin reviewers' Slack channel to maintain visibility across all uses of Devin for shipping, tagging in the appropriate engineer or designer for each PR.
Go find another team in your company and build a tool for somebody in a different org to make their life better as a first AI step.
Pick up one part of your workflow that is taking a lot of time and automate it, such as having an agent prep you before a customer call by pulling from multiple sources.
Run a hackathon where everyone participates to change the idea that you have to be technical to build something.
As a marketing leader, color code everything in your work map that should be AI enabled — for example, copywriting should not be written by hand, and AI video tooling should replace studio spend.
In interviews, ask candidates to screen share and show live how they use AI to quickly determine whether they are at Level 1 or higher.
Check out her class at Stanford if you are in the Bay Area so you can really learn AI PM. And if you are a leader, check out her course at Reforge.
do check out my bundle at bundle.akashsharma.com to get access to nine AI products for an entire year for free. This includes Dovetail, Mobbin, Linear, Reforge, Build, Descript, and many other amazing tools
“you should be spending time building the agent to pull the competitive data, and you should just be moderating it, but you shouldn't actually be doing the deep work every single day.”
“now that you can build faster than ever before, it doesn't mean you just build everything. Like what actually is important is to know why and for whom you're building for, and what is it that you're trying to solve for, and what success looks like”
Shared with Earmark · earmark-ai.com