Learn the frameworks reshaping the org chart, the skills that keep operators essential, and where to start.
This piece is adapted from an Operators Guild Focus Session on the AI-native operator, featuring Lawrence Coburn and Taylor McLoughlin, founding partners at Legible.co and co-founders of Ambient, and shaped by a live discussion among operators who are working out what their role becomes as AI moves from a bolt-on to the center of how companies run.
Focus Sessions are small-group, member-only conversations where operators compare notes on decisions in flight, pressure-test tradeoffs, and surface the operational realities that rarely show up in polished frameworks.
If you want access to sessions like this, including the recordings and the community behind them, you can apply to join OG.
The first wave of AI at work helped people move faster inside the tools they already used. The next wave is reshaping the company itself: the org chart, the operating model, and the role of the operator inside it.
That shift is early. There are no AI-native companies yet. No Apple, Google, Microsoft, Amazon, or Meta playbook exists for an AI-native company at scale, and there are no AI-native knowledge workers either. Everyone in the workforce today knew the world before AI. The story is being written right now.
So the useful question is not what the future looks like. It is what direction the evidence points, and what an operator should do about it now.
What "AI-native" actually means
A simple test cuts through the noise: if you were setting up your company today, how would you set it up to make the best use of AI?
That question forces hundreds of decisions that are hard to change later:
- How would you structure the org and set the rhythm of business?
- What tools would you choose, and would you reject any that lack an MCP connector?
- Who would set your recording policy, your lawyers or your CEO?
- How would you name your documents so a model can find them?
- What would your communication norms and values be?
The same test applies to a career. If you were starting yours today, which skills would you build? The answers cluster around loop design, context engineering, judgment, taste, systems thinking, domain depth, and model agility.
Transparency becomes more valuable in this frame, because AI needs context. The company that writes things down, records its meetings, and keeps its knowledge findable gives its tools something to work with. The company that keeps its best thinking in hallways and text threads keeps its AI half-blind.
Three frameworks shaping the AI-native company
The people with the best access are converging on a few ideas. Three frameworks are worth knowing.
The collapsing org chart
Jack Dorsey and Roelof Botha's work out of Block and Sequoia is likely to be read as the defining piece of this era. As Dorsey puts it, "I don't think this is a productivity thing. I think it's a structural thing."
His core claim is that hierarchies existed to route information. A leader capped out around eight reports because that was the limit of what one person could compile and pass up, or communicate down. AI does that routing now, so the layers that exist mainly to move information up and down are the exposed ones. Everything collapses toward three roles, which we return to at the end: builders, directly responsible individuals, and player coaches, with humans living at the edge where intelligence meets reality.
The prerequisite is legibility. If your company is not legible, you cannot put intelligence on top of it.
This is already happening. Brian Armstrong's org memo at Coinbase set a ceiling of five layers below the CEO, with no pure managers and AI-native pods.
Closed loops
The second framework, from Diana Hu at Y Combinator, treats AI as the operating system, not a tool. Her instruction is blunt: "You cannot outsource your conviction. Sit with the agents until they break your priors."
Most people use AI in an open loop. You feed it context, it produces an output, the thread ends. A closed loop feeds the result back in so the system improves. The newsletter Every captured the shape of it: inputs to a model to outputs, with a signal from the output routed back into the inputs.
Picture a thermostat that senses the room and adjusts against a target, next to a space heater that runs until someone shuts it off. With the space heater, you are the loop.
Three loops most operators already touch show the difference:
- Reporting. Open: AI drafts the report, finance fixes it, next month it repeats the same mistakes. Closed: AI remembers the fixes and which issues mattered, so next month's first draft is better.
- Board prep. Open: AI helps build the deck, then the board's questions and decisions get lost in the notes. Closed: AI captures what the board cared about and what management promised, and the next deck starts there.
- Vendor management. Open: AI warns you about a renewal but never learns whether you saved money or improved the service. Closed: AI tracks the deal, usage, and service afterward, and uses those results to prepare for the next renewal.
Hu's related principle, "token maxing, not headcount maxing," says to put models to work before hiring armies of people. The follow-on lesson, learned the hard way by teams that burned a year of tokens in a quarter, is that token efficiency now matters as much as token appetite.
The era of mass cognition
The third framework, from Ann Miura-Ko at Floodgate, comes out of a tour of the closest things to AI-native companies. Her thesis is that software stops being a tool and becomes a collaborator.
Being "red-pilled" on AI is a scale, not a switch, and she measures it with four questions:
- What can AI see? Does it have full access to the knowledge of the business?
- What can it do? What can it complete without a human in the loop?
- Who can extend the system? Who widens its access and capability?
- How has the org changed? Have you rebuilt the workflow around AI, or just supercharged the old one?
Floodgate's goal is self-driving operations that sense reality, diagnose, initiate work, update memory, and change the company. Her warning is to beware the feature factory: when you can build anything, what you choose not to build is what matters.
The through-line: make your company legible
Across all three frameworks, one idea repeats. For AI to help a company, the company has to be legible. Its knowledge has to live somewhere a model can reach.
That is a bigger job than recording meetings. It means:
- documents named so they can be pulled into the right context
- meetings used to get thinking out of people's heads, not to read out status
- knowledge memorialized rather than shared in a hallway or over text
A poorly named document might as well not exist, because it cannot be pulled into context for the right project or customer. No company has this fully solved, which is exactly why it is an opening. Legibility is a documentation game, and operators are well placed to lead it.
What the most AI-native companies do differently
Drawn from more than 500 leadership interviews, six traits show up again and again. Three are about the stack, three about the org.
- A ranked AI roadmap. Automation is run as a program with owners and dates, not a pile of demos. Every key workflow is stack-ranked by impact and hours consumed, at the company level and inside each team, and progress is reviewed on the same cadence as revenue.
- Team gains, not just individual gains. Individual productivity is table stakes, with 82% of tech workers already reporting AI productivity gains. The compounding now happens at the team layer: a library of company skills with named owners, MCP enabled for key systems, and a shared context layer that lives outside the model on Notion or Drive.
- No stragglers. The gap between the best and worst AI user is small because enablement is for everyone. Training is tied to real workflows rather than generic tool tours, leadership models AI use visibly, and internal superusers teach instead of only outperforming.
- An operating model that got the memo. Norms change to match the stack. Meetings are for debate and ideation, status is an artifact AI produces, and data hygiene and naming conventions are owned at the leadership level.
- Token efficiency. The bill is high on purpose, and none of it is wasted. Written token governance routes frontier models to judgment and small fast models to volume, batches what is not latency-sensitive, and reviews spend like any other strategic budget.
- Not married to a single model. Capabilities leapfrog every few months, so the best companies treat the model as a swappable engine. Context lives in portable form, work moves across Claude and ChatGPT as capabilities shift, and switching cost is measured in days. MCP going from one vendor's project to a cross-vendor standard in under a year made portability an architecture choice.
The layers operators should think about
A helpful way to organize the work is by what you make reusable for the team, from lightest to heaviest:
- Skills: reusable instructions for a task or workflow, like a branding skill that keeps every deck on-brand without anyone thinking about it.
- Plugins: a level deeper, bundling multiple skills and resources into something more integrated.
- Agents: automations that run work on their own, with the harder question of how to run them in the cloud rather than on one person's machine.
- The brain: the context layer itself, where transcripts and documents live in a form that agents and humans can both use.
Each layer adds capability and adds requirements for access, security, and ownership. The biggest shift happens when a personal tool becomes something the team depends on.
The AI-native operator's core skills
If orgs are flattening, knowledge is moving into the machine, and context is everything, three skills hold their value.
Build, even if you are not an engineer
No role in the modern org is untouched by building, and that goes up to the CEO. Building does not mean shipping production code. It means using the tools instead of working around them. Turn a strong chat thread into a skill you can reuse. Automate the paper cuts. Go one exercise past your comfort level, even if that means coding a game over the weekend to prove you can.
The posture to avoid is the executive who has an assistant print out every email to handwrite the replies. If you are not taking advantage of the tools available to you, you are a dinosaur.
Deepen your domain expertise
AI gets you to a pretty good answer on almost anything, roughly eighty percent of the way. The value is now in the last mile, the taste and judgment that carry a good output to a great one. That window is narrowing, which is why it is worth protecting.
This is also why AI is not only a young person's game. Twenty-five years of taste in a domain is a real asset when the job becomes being the last, best set of eyes on the work. The era of the pure generalist adding outsized value is closing, because that is the layer AI covers first. Whatever your area of depth, keep deepening it.
Think in systems, not tasks
As AI takes on more individual tasks, the value moves to coordination. How do you organize the work and point multiple agents at pieces that have to fit together and add up to a goal? Engineers already orchestrate agents that build features in concert. The business side has the same shape. It looks a lot like classic management, which is being very good at coordinating the thing rather than doing the thing.
Where you fit: three roles survive the flattening
Place your current role against the three roles Dorsey expects to last.
- Builder. People who ship. With AI leverage, one operator can now build the automations, reports, and tools a team of five used to. The question to ask: what did you make this month?
- Owner, the directly responsible individual. Someone accountable for a result end to end, not a task list, often a share of the P&L. Deep knowledge of the ICP and the product still matters. The question to ask: what outcome has your name on it?
- Player coach. A leader who sets context and standards while still doing the work. Pure management, the routing job, is the role that disappears. The question to ask: what did you do, not just review?
Where to start
Operators have a real opening to lead their companies through the transition. A few moves come first:
- Lead the push toward legibility: findable data, a clear recording policy, meetings used for thinking, a culture of writing things down.
- Act as a change agent, modeling the behavior you want from the team, since change starts with executives using the tools themselves.
- Pull up the stragglers by protecting budget for training and enablement.
Then run the honest audit. Find where the hours are going and where the dollars come from, rank the top three workflows, and go after those with AI. The people already know where their time goes. Whether you are a nonprofit or a for-profit, a 15-person team or an 80-person association supporting 50,000 members, the math is the same: give hours back to your people so a smaller team can do the work of a larger one.
None of this is settled. The best-positioned people are still debating full automation, token budgets, and which model leads next quarter. What holds is steady. Orgs are getting flatter, knowledge has to be memorialized, and context is everything. The operators who lead their companies toward that, and who keep building, are the ones defining the AI-native operator while the rest of the market waits to find out.
Join the conversations operators are having now
This piece was adapted from an Operators Guild Focus Session, where senior operators compared frameworks, real workflows, and lessons from putting AI to work inside their companies.
OG members get access to Focus Sessions like this one, including the live discussion, recordings, and the community continuing the conversation afterward.
Apply to join Operators Guild and be part of the room.