On a Monday morning in a typical small business, sales might be using AI to clean up proposals, support might be trialling an inbox assistant, marketing might be drafting content faster, and operations might be testing automations in the background. Nothing about that sounds irresponsible. In fact, it probably sounds sensible.

The problem arrives later, when the founder or business leader cannot answer three basic questions: what changed for the customer, which workflow actually improved, and who owns the next decision?

That is where AI starts to expose the organisation. Not because the tools are bad, but because the operating model has not caught up. AI does not become valuable when more people use more tools. It becomes valuable when leaders embrace a new mindset, turn experimentation into owned workflows, build better customer outcomes, and embrace continuous learning.

That, for me, is what founder mode means in the age of AI.

What is AI Founder Mode?

AI Founder Mode means business leaders stay close to the decisions that shape AI adoption: which workflows matter, who owns them, where human judgment is required, what risks are acceptable, and how success will be measured. It is not micromanagement. It is leadership responsibility in a period where AI is changing how work gets done.

It does not mean doing everything personally. It does not mean becoming a bottleneck. It does not mean wandering around the business interfering in every decision. It means staying close enough to the work to understand where AI should matter, which workflows deserve priority, where human judgment still matters, and how success will be judged.

AI can create useful momentum quickly. It can also spread confusion quickly. If leadership steps too far away too early, the business usually gets scattered pilots, vague accountability, and more AI activity than operating change.

Why founder mode matters now

The phrase “founder mode” became popular after Paul Graham wrote about the difference between founder mode and manager mode. His basic point was that the standard advice given to founders as companies scale is often to hire good people, delegate to them, and stay out of the details. Graham argued that this can be broken advice for founders, because some of the most important work of leadership requires staying closer to the company than conventional management theory allows. 

The danger, of course, is obvious. Founder mode can be misused. It can become an excuse for ego, control, impulse or micromanagement. But the useful version is not that. The useful version is presence: a leader staying close to the few decisions that define how the business actually works.

That matters even more with AI because the workflows themselves are changing. We are not just choosing software. We are deciding how work gets done, what judgment gets delegated, what gets automated, what gets reviewed, and what kind of experience the customer receives.

Brian Chesky, Airbnb and the cost of organisational distance

Brian Chesky’s story at Airbnb is helpful here, and his story was the spark for Graham’s essay. During COVID Chesky effectively took back direct operating control of AirBnB having previously followed the advice to “hire good people and give them room to do their jobs.” Graham points out that ‘he followed this advice and the results were disastrous.’ On 5th May 2020, Airbnb announced that global travel had come to a standstill, that revenue was expected to fall sharply, and that nearly 1,900 of its 7,500 employees would leave the company. Airbnb also paused Transportation and Airbnb Studios, scaled back Hotels and Lux, and refocused around its core hosts, homes, experiences, belonging and human connection.

The deeper issue was not just cost. COVID exposed complexity. Airbnb had become too spread out. Different divisions were pursuing different bets. Work was duplicated. Accountability had slowed. Product direction had become less coherent. Chesky’s conclusion was that Airbnb could no longer afford to do everything. He had to make the company smaller, simpler and more focused.

What he appears to have discovered through that reset was that the standard executive delegation model had created too much distance between leaders and the details that mattered. Leaders had become too far from the product, the customer, the data and the real trade-offs. That is the version of founder mode that interests me: not the mythology of the heroic founder, but the operating reality of a leader who realises that scale has created abstraction.

By early 2026, Chesky was applying the same lesson to the AI era. In an Invest Like the Best podcast episode titled: “AI Founder Mode”, Chesky discussed why AI founder mode will demand even more attention to the details. 

Chesky explored how he didn’t want Airbnb to be a one-hit wonder. Airbnb has built one of the great consumer companies of the last twenty years, but the next question is whether it can build multiple new lines of business. That requires a different operating model: smaller teams, more direct judgment, less abstraction, and more leaders who understand the work rather than just the reporting lines. As a hands-on founder, Chesky then spends time with each of these teams. That must be hugely re-invigorating again. 

I say that with some affection and a little scar tissue. I once went up against Airbnb when it was first being born, and I obviously lost. So I watch Chesky with interest. Not as a fanboy, but as someone who knows how powerful clarity, timing and execution can be when they come together. I am cheering him on because I think the next phase will be fascinating. 

Jack Dorsey, Block and the intelligence layer

Jack Dorsey has been making a similar, and in some ways more radical, argument at Block. In early 2026, Block announced that it was cutting more than 4,000 employees, with Dorsey explicitly linking the restructuring to what AI now makes possible. The direction of travel appears to be a smaller, faster organisation built around new forms of AI-enabled coordination. (AP News)

That is a hard message, and I do not think we should skate over the human cost of it. Thousands of people losing their jobs is not a neat management theory. These are real people, real families, real consequences.

But the structural argument matters. Dorsey’s view, as I read it, is that hierarchy has historically existed to solve a problem: information flow. Companies got far bigger than any one person could understand, so they created layers of managers to route information up, down and across the organisation.

Block is now asking whether AI can perform more of that coordination function. In its own framing, Block wants to move from hierarchy to intelligence. The company argues that AI can maintain a continuously updated model of the business and coordinate work in ways that previously required people relaying information through layers of management. (AP News)

That is a profound shift. Chesky’s version is founder-led product intensity. Dorsey’s version appears to be organisation design rebuilt around an intelligence layer. Both point in the same direction: AI is collapsing the distance between leadership, work and customer reality.

That does not mean every business should copy Airbnb or Block. Most small businesses should not. You are not Airbnb. You are not Block. You do not need a grand theory of the future corporation before making a useful decision this week. But you do need to pay attention to the pattern. AI does not just make individuals faster. It changes how information moves, how decisions are made, how work is coordinated, and how quickly weak operating habits become visible.

Continuous looping: the technical expression of AI Founder Mode

Recently I watched Y Combinator partner Diana Hu unpack the idea of a closed loop, and it helped crystallise why AI Founder Mode is not just about using better tools. It is about changing the operating system of the company itself.

Her point, was that AI-native companies will not run like traditional companies with AI bolted on top. They will run as closed-loop systems. Every decision, customer interaction, code commit, design choice and product outcome becomes part of a live feedback system. The company does not just do work and report on it later. It captures what happened, learns from it, and feeds that learning back into the next cycle of action.

That is very close to the intelligence layer Dorsey describes at Block. Most companies still operate in open loops. Work happens in meetings, Slack threads, documents, dashboards and people’s heads. Context gets lost. Leaders depend on layers of managers to summarise what is going on. By the time information reaches the top, it is often delayed, simplified or distorted.

That is a huge source of frustration in any growing company. You either discover that multiple people have been working on the same thing without knowing it, or you create a culture of over-reporting to stop that from happening. Either way, the organisation pays a tax in duplicated effort, slower decisions, more meetings, and the constant feeling that everyone is busy but not always aligned.

Continuous looping changes that. The organisation starts producing machine-readable artifacts that AI agents can query, reason over and act on. Everything is recorded. The obvious examples would be internal meetings, or sales and customer service calls so instead of waiting for a status update, an AI system can understand the current state of the product, the customer, the codebase, the roadmap and the business in real time.

That does not remove the need for leadership. It changes what leadership is for. You need fewer people whose primary job is to collect updates, translate context and route information. You need more leaders who are close to the work, understand the details, and can use AI to improve the speed and quality of decision-making.

For me, continuous looping is the technical expression of AI Founder Mode. It turns the company from a hierarchy of people passing work around into a live system that can sense, act, learn and improve. The goal is not simply for the founder to do more. It is for reality to reach the founder faster, and for execution to compound with every loop.

Manager mode for AI versus AI Founder Mode

The difference is not philosophical. It is operational.

Manager mode for AI	AI Founder Mode
Treat AI as an innovation project	Treat AI as an operating model shift
Start with tools	Start with bottlenecks, customer friction and decision latency
Measure activity and usage	Measure cycle time, decision quality and duplicate work removed
Add AI to existing workflows	Redesign workflows around continuous loops
Rely on managers for status updates	Build systems that make the company queryable in real time
Accept context living in meetings, Slack, docs and people’s heads	Turn work into machine-readable artifacts that agents can reason over
Create generic AI guidance	Set workflow standards, review points and guardrails
Run scattered experiments	Prioritise a few high-value use cases that change how work gets done
Use AI to produce more outputs	Use AI to improve the next cycle of action
Preserve the pyramid and make it more efficient	Flatten parts of the organisation around an intelligence layer
Reward coordination and reporting	Reward ownership, judgment and proximity to the work
Talk about urgency	Convert urgency into owners, deadlines, loops and operating rules

The key shift is this: not, “Are people using AI?” but, “Is AI changing how the company senses, decides, acts and learns?” That is the founder mode distinction.

Why AI exposes weak operating habits so quickly

AI exposes weak operating habits faster than most change efforts because it is easy to start and hard to integrate. A team can open an account, test a tool, or automate a step in an afternoon. That low barrier is useful. It is also dangerous.

When something is cheap to begin, broad in impact and difficult to govern once embedded, leadership discipline matters more, not less.

The evidence points to exactly this gap. Goldman Sachs 10,000 Small Businesses Voices reported that 76% of small businesses were using AI, and 93% of those users said it had a positive impact. But only 14% said AI was fully integrated into core operations, while 73% said more training and resources would help them implement AI successfully. (Goldman Sachs)

That does not prove weak leadership is the only cause. But it does show the core problem. In small businesses, AI use is running ahead of AI integration. Broader market data points in the same direction. Nasdaq’s 2026 Outlook Survey found that only 5% of leaders described their organisations as AI-native, while 71% categorised themselves as AI “newbies” or “explorers”. (Nasdaq)

WRITER’s 2026 AI adoption research found that 75% of executives admitted their AI strategy was more for show than actual internal guidance, while only 29% reported significant returns from AI. Those broader sources are not SME-only evidence, so they should be read carefully. But they reinforce the same pattern: many organisations have AI activity; far fewer have AI operating discipline. (WRITER)

Proof snapshot

The pattern is clear: AI activity is running ahead of AI integration.

1. 76% of small businesses report using AI, but only 14% say it is fully integrated into core operations. Source: Goldman Sachs.
2. 5% of leaders say their organisations are AI-native. Source: Nasdaq.
3. 75% of executives say their AI strategy is more for show than actual guidance. Source: WRITER.

That is why AI makes weak leadership visible so quickly. It crosses every function. It changes judgment, not just speed. It reaches the customer quickly. And it creates the illusion of progress, because tool usage can feel like momentum even when no core process has improved.

Five signs your AI effort is being delegated into failure

If you want a blunt diagnostic, start here.

1. Nobody owns a specific bottleneck.
   People are “using AI”, but nobody owns a concrete business problem such as lead response time, proposal turnaround, support triage, onboarding admin or internal knowledge retrieval. Activity is everywhere. Accountability is nowhere.
2. The founder hears about AI in updates, not in decisions.
   Leadership gets summaries after experiments have happened instead of helping decide where AI should matter first, what success looks like, and what risks need control. The founder does not need to approve every prompt, but they do need to be close to the decisions that shape the operating model.
3. Usage is broad, but workflow change is shallow.
   Teams are drafting faster, summarising faster and producing more outputs. But service levels, review steps, handoffs and operating procedures still look the same. The tool changed. The business did not.
4. Governance is either absent or performative.
   Some companies have no guardrails. Others have a generic AI policy that nobody can apply to real work. Both create the same problem: unclear responsibility when the output matters.
5. You cannot explain the customer impact.
   If the business can say it is using AI but cannot say whether customers are getting faster answers, clearer communication, or fewer avoidable errors, the effort is drifting. Customers do not care whether a team is “exploring AI”. They care whether the business is easier to deal with, faster to respond, and more reliable when it counts.

How to diagnose the AI ownership gap

If AI activity is rising but execution still feels scattered, I would start by looking for the gap between enthusiasm and execution. Usually, the problem is not that people are unwilling to try. It is that nobody has clearly decided which business problem AI is meant to solve, who owns the workflow, and what better actually looks like.

This is where I would help a client slow the conversation down. Before adding more agents or launching another pilot, I would want to identify what is already being used, where risk is emerging, where duplication is happening, which workflows are most ready, and which customer moments would benefit first.

The useful first step is not always implementation. Often, it is an audit. Where are you now? What are people already doing? What is working? What is happening unofficially? Where is the business exposed? Where is the customer feeling friction? Where would AI create the most useful improvement?

An audit gives you an independent view of where you actually stand. It creates clarity before momentum turns into sprawl.

A practical AI Founder Mode checklist for SME leaders

Founder mode does not mean personally controlling everything. It means staying close to the decisions that define how the business actually runs.

Here is how I would help a client think through it.

1. Pick one to three bottlenecks, not twenty tools.
   Start where time, quality, margin pressure or customer friction is already obvious. Do not begin with, “Which AI tools should we use?” Begin with, “Where is the business already feeling pain?”
2. Tie every AI effort to a business outcome.
   Faster quoting. Cleaner follow-up. Better knowledge retrieval. Fewer support delays. More consistent delivery. If we cannot name the business result, the use case probably is not ready.
3. Name an owner for each workflow.
   Someone has to own the process, not just the software. Otherwise AI becomes everyone’s experiment and nobody’s responsibility.
4. Set a founder review cadence.
   Weekly or fortnightly is usually enough in an SME. I would review live outputs and actual workflow changes, not slide decks about potential. What changed? What broke? What did the customer notice? What should we keep? What should we stop?
5. Define what is assisted, what is automated, and what stays human-led.
   Some work should be AI-assisted. Some work can be automated. Factor in the cost of tokens. Some work should stay human-led because the judgment, empathy, risk or trust requirement is too high. Teams move faster when those distinctions are clear.
6. Create simple guardrails for privacy, quality and escalation.
   Make them specific to the workflow. Generic policy language is rarely enough. People need to know what data cannot be entered, which outputs require review, and when a human must step in.
7. Train the team where usage is expected.
   “Use AI more” is not a training plan. People need to know where it fits, what good looks like, and when to stop and escalate.
8. Kill shallow experiments quickly.
   If a use case is interesting but not improving a meaningful process, I would stop it before it becomes another source of noise. Curiosity is good. Unbounded experimentation is not.
9. Document what good looks like.
   One repeatable workflow is worth more than ten vague experiments. If something works, turn it into a standard. Name the owner. Capture the prompt or process. Define the review point. Show the team. Improve it again.

For example, a small professional services firm might start with proposal turnaround. The current process takes three days, depends on one senior person, and often creates delays for prospective clients. A useful AI implementation would not simply “help people write proposals faster.” It would create a clearer workflow: intake notes captured in a standard format, a first draft generated from approved language, human review for pricing and judgment, and a final quality check before it goes to the client. The business outcome is not “we used AI.” It is that proposal turnaround moves from three days to the same day, without lowering quality or increasing risk. And before long the pricing guardrails enable the proposal to work end to end, and the final review is quick until it’s no more.  

This is founder mode in practice. Not ego. Not chaos. Responsibility.

I co-wrote a book called Futureproof with Minter Dial and Responsibility is one of the core mindsets we argue futureproof leaders need to develop. Not responsibility as a vague leadership value, but responsibility as a willingness to stay close to the real work, make clear decisions, own the consequences, and build the conditions for others to succeed.

That is what AI is now testing: not whether leaders can talk about transformation, but whether they can take responsibility for how work actually changes.

What changes when leaders stay directly engaged?

1. Customer outcomes improve first

When leaders stay close to where AI touches the customer, they make better judgment calls. Where will customers value speed? Where will they notice sloppiness? Where should a human still step in?

That changes the quality of adoption. AI stops being a pile of internal experiments and starts improving the things customers actually feel: faster responses, clearer communication, and fewer avoidable errors.

I felt this sharply today. I found myself stuck with a chatbot from a company I have been a customer of for 15 years. Having experienced brilliantly effective chatbots where the business owners have clearly spent time designing the experience properly, this one made my opinion of the company dip.

It was inefficient. It was taking up more time than it saved. And the more it struggled to help, the more aware I became of how precious that time was.

Theo Paphitis, a Dragon on the British TV show Dragons’ Den, used to ask when deciding whether to invest in a company: “Why should I invest my children’s inheritance into your business?” Today, I found myself asking the customer version of that question: why am I being asked to invest time I would rather spend with my children?

That is the customer test. If your chatbot genuinely helps, brilliant. If it cannot, the best thing it can do is move me quickly and gracefully to a human. Good automation should respect the customer’s time. It should make the experience feel easier, not leave people doing the work the system was supposed to reduce.

2. Processes become clearer

Weak execution creates tool sprawl. One team uses one assistant. Another builds a workaround. Another writes its own prompts. Nobody knows which version of “best practice” is safe or effective.

Direct engagement cuts through that. The founder does not need to own every step, but they do need to force a few hard decisions: which workflows matter, who owns them, what standards apply, and how results are reviewed. That is how experimentation becomes an operating discipline.

3. Innovation becomes cumulative

Many businesses already have enough curiosity. What they lack is a learning loop. A founder-led approach compounds learning. One useful workflow gets tested, improved, documented and scaled. One customer lesson shapes the next decision. One success creates a clearer model for the team.

Without that loop, innovation stays scattered. With it, the business gets better at knowing what is worth keeping.

What AI guardrails do small businesses need?

There is a bad version of founder mode that confuses urgency with impulse. Avoid it.

Responsible AI should be practical, not theatrical. The goal is not to create a policy that sounds impressive but never shapes behaviour. The goal is to give people simple rules they can apply when real work is moving quickly. Some will come to conclusions that the 

1. Ethics.
   What will we not use AI for? Where do we need disclosure? Where must a human stay in the loop?
2. Governance.
   Who can approve new AI use cases? What needs review before it becomes part of the workflow? Who decides when an experiment becomes a standard process?
3. Privacy.
   What information should never be entered into third-party tools? Customer data, contracts, financials, employee information and commercially sensitive material all need clear rules.
4. Quality control.
   Which outputs can be used immediately? Which needs review before they reach a customer, colleague or system? Where does speed create risk?
5. Role clarity.
   Who sets direction? Who owns the workflow? Who checks quality? Who supports implementation?
6. False urgency control.
   Not every process needs AI now. Prioritise where the customer problem, cost pressure or execution bottleneck is real.

Direct engagement is not the opposite of governance. In a small business, it is often how governance becomes real. How does the founder ensure speed does not outrun judgment?

When should you bring in outside AI help?

I think there are two good moments to bring in outside help.

The first is at the beginning. Wherever you are in the AI journey, an audit is useful because it gives you an independent view of where you actually sit. What tools are already being used? Where is the risk? Where is the duplication? Which workflows are most ready? Which customer moments would benefit first? Before you rush into implementation, it helps to get clarity.

The second is when the business has done enough to feel the friction. The team is experimenting, but results are uneven. Leaders can feel momentum, but cannot prioritise what matters most. Customer-facing use cases feel promising, but risky. Tool usage is growing faster than operating discipline. The founder knows AI matters, but does not yet have a clean execution model.

But the biggest early work is not technical. It is the leadership mindset and culture. There has to be curiosity in the room. Leaders need to try the tools, initially to get onto Claude or ChatGPT, and start asking better questions. What would make the customer experience easier? What work is taking too long? Where are we making people wait? Where are we asking staff to repeat low-value tasks? Where could AI help people do better work, not just faster work?

I also understand why some leaders hesitate. The human resource question is real, and fears about job losses need to be taken seriously. But there is another path: equipping people to use AI to increase the impact of their work.

Gary Vaynerchuk has made this point in his own way, arguing that AI should not simply be about replacing people, but about helping strong teams become more valuable and productive with AI. He has also said AI should be used to retain strong talent and increase revenue with roughly the same overhead, rather than only to cut roles. (garyvee.substack.com)

That is a different starting point, and one that businesses focused on their triple bottom line of people, planet and profit will be remembered for. Helping staff adopt AI is not only useful for the business now. It also equips the workforce for the future. The question is not just, “What can we automate?” It is, “How do we help people become more capable, more confident, and more valuable in the work that remains human?”

That is the practical role of outside help: not to admire the trend, but to audit where you are, work with the leadership team on mindset, identify where operating gaps are showing up, see where pilots are stalling, and help decide what should happen next. Sometimes that means giving AI a proper kickstart on the right foundations. Sometimes it means bringing discipline to work that has already started. Either way, the goal is the same: move from scattered experimentation to responsible, useful, customer-centred implementation.

FAQ: founder mode, AI adoption and small business implementation

1. What does AI Founder Mode mean?

AI Founder Mode means leaders stay close to the decisions that shape AI adoption, including workflow priorities, ownership, customer impact, risk and review standards.

2. Is Founder Mode just micromanagement?

No. Micromanagement pulls leaders into every decision. Founder mode keeps leaders close to the few decisions that matter most: priorities, customer impact, ownership and guardrails.

3. Why do AI pilots stall in small businesses?

AI pilots often stall because usage is easier than integration. Teams start using tools, but workflows, ownership, standards and review rhythms do not change.

4. Where should a small business start with AI?

Start with one to three bottlenecks, not twenty tools. Look for places where time, quality, cost or customer frustration is already obvious.

5. Do small businesses need to become AI-native?

No. That is not the first goal. The first goal is to make AI useful inside real work. Clear priorities, named owners, simple standards and regular review rhythms matter more than the label.

6. What AI decisions should founders own personally?

Founders should own the direction: which bottlenecks matter, where customers will feel the change, what risks are unacceptable, who owns the workflow, and how results will be reviewed. They do not need to own every task.

7. What AI guardrails do small businesses need?

Small businesses need practical guardrails for privacy, quality control, disclosure, human review, role clarity and escalation.

8. When should a business bring in outside AI help?

At the beginning, to audit where you stand and identify the best place to start. Then again when experimentation is happening, but clarity is missing. If AI use is growing and you still cannot point to cleaner workflows, stronger ownership or better customer outcomes, it is time to get help turning activity into execution.

Key takeaways

1. AI activity is not the same as AI integration.
   A business can have lots of people using tools without changing how work gets done.
2. Founder mode for AI is about responsibility, not micromanagement.
   Leaders need to stay close to priorities, customer impact, workflow design and guardrails.
3. Small businesses should start with bottlenecks, not tools.
   The best AI use cases usually begin where time, quality, cost or customer friction is already visible.
4. Responsible AI needs practical guardrails.
   Policies only matter if people can apply them inside daily work.
5. The goal is operating change.
   AI should lead to clearer workflows, better customer outcomes and faster learning loops.

A simple PIE check before your next AI meeting

Use this before your next AI meeting.

1. Personal.
   What do you, as the leader, need to understand firsthand rather than through a second-hand update?
2. Internal.
   Which workflow needs one clear owner, one clear standard and one clear review rhythm?
3. External.
   Where will the customer notice the difference first, for better or worse?

If those answers are fuzzy, the business is probably moving faster in tooling than in leadership. Your first move this week is simple: choose one workflow, one owner and one review date. That will teach you more than another month of vague experimentation.

The real question is not whether you are using AI

Most businesses no longer need convincing that AI exists. The harder question is whether someone senior is close enough to the work to turn experimentation into a better way of operating.

That is the real Founder Mode question. Not: are we trying AI? But: who owns the changes AI is making to our decisions, workflows and customer experience?

If the answer is vague, that is the risk. Not slow tooling. Slow responsibility.

Start with an AI ownership audit

If your business has started using AI but has not yet turned that activity into clear operating gains, start with an AI ownership audit.

We can help you identify what tools are already being used, where risk or duplication is emerging, which workflows deserve priority, who should own them, and what guardrails need to be in place.

From there, an AI implementation audit can turn scattered activity into a practical execution plan: one workflow, one owner, one review rhythm, and a clearer path to better customer outcomes.

Not another tool people are trying. A better way for the business to work.

Source note

This reflection draws on Paul Graham’s “Founder Mode” essay for the founder mode and manager mode distinction; Airbnb’s May 2020 reset and later comments from Brian Chesky; Block’s restructuring and “hierarchy to intelligence” framing; Goldman Sachs 10,000 Small Businesses Voices data on SME AI adoption; Nasdaq’s 2026 Outlook Survey; WRITER’s AI adoption research; and Gary Vaynerchuk’s comments on AI, workforce capability and people-centred adoptio

n.