AI-Native vs AI-First Company Culture
Reading Time: 7 Minutes
Image created by the Superhumxn team.
Billion-Dollar Valuations
In September 2024, Amazon's Andy Jassy told every team to increase its ratio of individual contributors to managers by at least 15%. His reasoning was that fewer managers means fewer layers, faster decisions, and less distance between the people closest to a problem and the people who can act on it.
Jassy is not alone. Bayer cut nearly half of its management and executive positions under a new model it calls Dynamic Shared Ownership, designed specifically to cut hierarchy and speed up decisions. It’s a safe bet that AI and automation are driving these changes. These industry leaders understand that the org chart most companies still rely on is now a liability.
What's replacing traditional company structure is an AI-first approach, where established businesses use AI to automate and streamline work. Essentially, they don't have much choice. These organisations are increasingly competing with a new generation of AI-native start-ups that were built around AI from day one. Many are reaching significant scale with teams of fewer than 50 people, while attracting billions in valuation and venture funding. Investors are paying close attention because these companies are proving that growth no longer requires layers of management or constant headcount expansion.
Gartner's fast work vs slow work framework maps which tasks AI can accelerate and which still require human judgment. It's a useful lens for understanding why AI-native companies are restructuring around fewer management layers, routing the fast work to AI and concentrating human effort where it matters most.
What can I expect from a subscription?
Real-world strategies for navigating the future of work.
AI AGENTS AT WORK
Everything you need to know about how AI agents are changing teams and ways of working.
SUPERHUMXN IRL
Real insights from investors and operators.
MEMOS ON PEOPLE, CULTURE & AI AT WORK
The latest use cases and strategies for change and transformation.
IRL
AI-First vs AI-Native
Microsoft's engineering team ran a three-month experiment putting AI agents alongside human engineers on an existing process. This is an AI-first approach in practice. Coding dropped from roughly half of engineering time to single digits, while strategic work, planning, architecture, review, jumped to 73%. In one exchange, AI agents reviewing a proposed chat interface caught a UX problem, a conversion problem, and a security vulnerability in the same conversation, all from one design choice.
Cursor is the AI-native version of the same idea. Built by Anysphere as a fork of VS Code, but rebuilt from scratch with AI "as a first-class citizen, not a sidebar," it doesn't bolt AI onto an existing workflow, instead the workflow is the AI. There's no separate step of writing code and then asking an assistant about it; the editor, the AI, and the codebase share context constantly.
The difference isn't really about which is "better." It's that AI-first companies are working out how to integrate something new into something that already works, while AI-native companies never had to unlearn the old way in the first place.
Microsoft's three-month experiment embedding AI agents into an engineering team found that human time shifted almost entirely away from writing code and toward reviewing, directing, and catching AI errors. It's one of the clearest pictures available of what AI-first engineering actually looks like day to day.
AI-Native Companies Communicate Differently
Many organisations still treat communication mainly as information exchange. In practice, communication systems directly affect how teams work day to day. Meetings affect how much uninterrupted focus employees have. Documentation affects how quickly new employees learn. Visibility across workflows affects trust between teams. Feedback timing affects whether employees feel comfortable raising problems early or stay quiet until issues become harder to solve.
These patterns become much more obvious inside distributed AI-native companies where teams cannot rely on office proximity to fill communication gaps. Once employees work across different locations, time zones, and systems, organisations need much clearer ways of sharing information and coordinating decisions.
This is one reason many AI-native companies place unusually high importance on written communication, transparent documentation, visible workflows, and asynchronous collaboration. Some companies deliberately reduce unnecessary meetings so employees have more uninterrupted focus time. Others make decision-making processes highly visible internally so employees can understand priorities, context, and ownership without waiting for multiple approval layers or clarification meetings.
The strongest teams often share one important characteristic: employees spend less time trying to figure out what is happening around them.
People know where information lives, how decisions get made, what priorities matter most, and how work connects across teams. That becomes increasingly important once employees are already working inside environments where tools, workflows, and priorities are changing constantly.
Image Source. As AI agents increasingly work alongside each other, not just alongside humans, the way they communicate and coordinate becomes a structural question, not just a technical one. The architecture choices made here will shape how AI-native companies actually operate at scale.
From Nvidia to Shopify, here's how ten companies are rebuilding their culture around AI, and what you can take from each.
AI-First
1. Jensen Huang - Nvidia
Tae Kim's book The Nvidia Way describes a culture built around speed, intellectual honesty, and an almost obsessive work ethic, all set by Huang himself. He has reportedly told executives, "There may be people smarter than me, but no one is ever going to work harder than me." Fear and anxiety are, by Kim's account, deliberate motivational tools Huang uses to keep the company moving fast and avoid the complacency that has sunk other tech giants.
What that looks like day to day: flat reporting lines (Huang has reportedly had dozens of direct reports at once), public real-time critique of work in meetings, and a relentless focus on staying paranoid about being overtaken. The lesson for other organisations is less about specific AI tools and more about what Huang has built deliberately: a culture where avoiding complacency is treated as the central job of leadership, not a side project.
AI-Native
2. Greg Brockman - OpenAI
OpenAI's culture has been shaped by an unusually fast and public iteration cycle. GPT-3, GPT-4, ChatGPT, GPT-4o, Sora, each shipped, got used by millions of people immediately, and was refined based on what happened next.
Greg Brockman and Sam Altman have described the approach openly: "We need to have a very tight feedback loop, rigorous testing, careful consideration at every step." There is no fixed playbook, by their own admission, so the company leans on continuous testing and feedback to navigate territory nobody has mapped before.
The lesson for other organisations isn't "move fast and break things." It's that when you're operating somewhere new, with no established best practice, a tight feedback loop becomes the substitute for a roadmap. You learn what works by shipping something small, watching what happens, and adjusting quickly, rather than trying to plan the whole thing upfront.
AI-First
3. Ali Ghodsi - Databricks
Databricks was ranked #2 on Glassdoor's inaugural Best-Led Companies list in 2024. The two values employees most often point to are truth-seeking and first principles thinking, both traced back to the company's origins as a research lab at UC Berkeley.
Ghodsi runs a weekly company-wide all-hands that opens with an "ask me anything" session, before any other agenda item. Employees can ask him anything, every week, with no pre-screening. That single recurring habit does more to signal what "transparency" actually means at Databricks than any values statement could.
The lesson here is that transparency as a value means nothing until it's an actual calendar event that leadership shows up to every single week, even when there's little to say.
AI-Native
4. Demis Hassabis - Google DeepMind
Hassabis has talked about what he calls a research closed-loop: a system where AI can ask questions, run experiments, verify results, and iterate on its own, without waiting for a human to manually design and run the next test. AlphaFold was the proof of concept. DeepMind is now building toward a fully automated laboratory, opening in 2026, designed from scratch around this idea.
What's notable about Hassabis's own leadership style is what his biographer Sebastian Mallaby found when he went looking for the catch. Hassabis told him: "The worst thing you can do to somebody is to be controlling. I go to great lengths not to be like that."
The throughline connecting both: build the system so that testing, verifying, and adjusting happens continuously and with minimal friction, whether that system is a lab full of researchers or an automated experimentation pipeline. Then get out of the way.
AI-First
5. Alexandr Wang - Scale AI / Meta
In June 2024, Wang formalised "MEI" at Scale: merit, excellence, and intelligence, explicitly positioned against DEI frameworks. "Hiring on merit will be a permanent policy at Scale," he wrote, arguing that a hiring process based purely on merit would naturally produce diversity as a byproduct, not a target.
The post was widely praised by figures like Elon Musk and Coinbase's Brian Armstrong, and equally widely debated. Research published in Administrative Science Quarterly found that when organisations explicitly promote meritocracy as a core value, managers can become more prone to bias in practice, not less, because believing your system is already fair makes you less likely to scrutinise your own decisions.
In June 2025, Wang left Scale to lead Meta's new Superintelligence Lab as part of a $14.3 billion investment deal.
For leaders, the lesson is about recognising that the principles you stand for will ultimately shape the culture you create.
AI-Native
6. Sarah Guo - Conviction
Sarah Guo publishes all of Conviction's LP letters publicly, essentially open-sourcing her own investment thinking, including the calls that didn't pan out. Her reasoning: "You are much more likely to find the truth if you are curious and willing to be wrong."
That principle runs through how she talks about AI itself. Guo has pointed out that in 2016 she considered building an AI language-learning startup with Andrew Ng and decided the technology wasn't ready yet. Looking back, she now believes the market was actually possible all along, the assumption that killed the idea was simply wrong. Her conclusion: in AI, "a lot of your priors, your existing beliefs about markets, just don't make sense anymore."
The lesson extends beyond investing. If your operating assumptions were formed even two or three years ago, in an AI context, treat them as provisional rather than settled, and build in a way that makes it easy to discover when they've stopped being true.
AI-Native
7. Tomasz Tunguz - Theory Ventures
Theory Ventures describes itself as an "investing corporation", structured so that researchers, engineers, and operators sit alongside the investors making decisions, with real-time market maps and in-house AI tooling built directly into how the firm works.
That's a meaningfully different structure from a typical VC firm, where research and analysis support investors but sit organisationally separate from them. Tunguz's bet, reflected in his blog at tomtunguz.com which draws millions of page views a month, is that the firms which build their own data and research capability internally, rather than treating it as a support function, will out-decide the ones that don't.
The takeaway is this: don't outsource or side-line a capability that's core to how you make decisions.
AI-First
8. Mustafa Suleyman - Microsoft AI
Suleyman's book The Coming Wave, published in 2023, argues that AI represents an unprecedented step change in human capability, and that the people building it carry a direct responsibility to confront the risks rather than defaulting to either blind optimism or paralysis. In March 2024, Microsoft hired Suleyman to lead its newly created Microsoft AI division, bringing much of his Inflection AI team with him, including co-founder Karen Simonyan.
The acquisition was effectively a culture transplant. Rather than building a consumer AI culture from scratch internally, Microsoft imported an existing team and the working philosophy that came with it.
The lesson here is structural. If a capability or culture you need doesn't exist and would take years to build organically, sometimes the faster path is acquisition.
AI-First
9. Tobi Lutke - Shopify
In April 2025, Tobi Lütke's internal memo declared "reflexive AI usage" a baseline expectation at Shopify. Teams now have to demonstrate that AI genuinely can't do a task before requesting headcount or additional resources, and AI fluency became a formal part of performance reviews.
What's easy to miss is that the memo wasn't really the start of anything. By the time Lütke sent it, Shopify had already built an internal LLM proxy and more than two dozen internal MCP servers, infrastructure that made "use AI for this" something employees could actually act on immediately rather than a lofty vision. The memo accelerated a shift that had been underway for years.
The takeaway is that a mandate like this only works if the infrastructure already exists to make compliance possible. Announcing "use AI more" to a workforce with no internal tools to use goes nowhere. Shopify's memo worked because the tools were already in place.
AI-First
10. Julia Liuson - Microsoft/GitHub
GitHub Copilot now writes around a quarter of the code in GitHub projects, and 90% of the Fortune 100 use it. In 2025, Julia Liuson, who leads Microsoft's developer tools division, sent a memo saying AI usage should factor into managers' "holistic reflections" on employee performance, not as a box-ticking metric, but as part of an ongoing conversation about what someone learned and how they're developing.
Then-GitHub CEO Thomas Dohmke publicly backed the approach, calling it "totally fair game" for 2025: did you use Copilot to summarise a meeting, and if not, why not?
Rather than enforcing AI use as a pass/fail metric, the lesson here is to build it into the kind of qualitative, ongoing conversation managers are already having with their teams about growth and learning.
Convergence
These leaders sit somewhere on a spectrum between AI-first and AI-native approaches. Essentially, their paths converge on the same outcome: a flatter organisational hierarchy with more autonomy distributed among teams. This is only a preview of what's to come.
