Four Moves That Decide Who Survives The Next 18 Months Of AIst

It's not that companies aren't using AI. It's that they can't.

Most of the frustration owners have with AI not doing what they want is actually them not being able to describe what they want. AI is execution. It is quite powerless when it does not know what to execute. The companies thriving with it right now are the ones who could already answer five questions in sixty seconds before AI showed up. What we sell. To whom. Why we win. What blocks us. What we ship this month.

If you cannot answer those, AI will not save you. It will help you flail more impressively. With more backflips and charts.

I have watched what happens when an owner can answer those five questions. One Strategy Sprints client, a professional services team in B2B, grew revenue 130% in 90 days and returned 144 times what they paid us. The model did not grow them. The discipline did. AI just made the discipline cheap.

Three things are about to decide who makes it to the next phase of this market. None of them are about model quality. A fourth move decides who keeps the customers they earn.

For two years the market competed on smart. Whose model was smartest, fastest, most reliable. The models are now good enough. The harder question for any company today is not whether the model can help. It is whether the company will let the model act inside real systems.

Permission here is not abstract access control. It is the right to write code into your repo, touch production, open and close tickets, change configurations, message customers, approve workflows, trigger downstream actions across your stack. Before that boundary the model advises. After it the model operates. Those are two different products.

Every platform shift in the last twenty years has followed the same loop. A company sells you a capability. The capability creates new governance, security, and operational burdens. The same company is then best positioned to sell you the layer that manages those burdens, because it has the deepest integration and the most complete view of how the system actually behaves in practice. Each side of the relationship reinforces the other.

You have lived this. The 200-person SaaS firm that moved its infrastructure to AWS years ago is now buying GuardDuty, Security Hub, and WAF, because best-of-breed alternatives cost more friction than they save. The 80-person consultancy that standardized on Microsoft Entra for identity is now buying Defender and Security Copilot, because Microsoft sees the most context and stitching another vendor in takes weeks of integration that nobody on the team has time for. The vertical SaaS company embedded in a hospital workflow keeps selling the same hospital its compliance and audit add-ons, because once a system is where the decisions happen, it also becomes where the decisions have to happen safely.

This is the loop. Capability creates burden. The same vendor sells the cure. Each side reinforces the other.

AI is the same loop, but compounded. The capability and the threat are the same artifact. The model that helps you patch a vulnerability is the same model that could exploit one. The model that drafts your client emails is the same model that could leak your pipeline. So the company governing the model is also the company that best understands what the model can do when it is turned against you. Capability layer and trust layer are arriving from the same vendors at the same time.

There is a second twist. Old infrastructure did not get smarter the longer you ran on it. AWS did not learn your business by hosting your workloads. Frontier AI does. The longer it operates inside your company, the more it understands how you actually run. Switching means rebuilding context from scratch. That lock-in is new.

Translation for a small B2B owner. Every permission you grant in the next twelve months will compound. You will not get the chance to thoughtfully review them later. The vendor that gets inside your inbox, your CRM, and your codebase first will know more about your operation than you do, and will be the only realistic option to govern that knowledge.

That is not a reason to wait. It is a reason to be the one in the driver's seat while the seat is still soft.

A computer needs direction. The single thing most owners are missing is not models or tools. It is clarity at the top.

I have consulted for the world's largest companies, hundreds of startups, and tons of mid-size firms in the Global 1000. The number one issue is unclear and constantly-changing vision and goals. A massive percentage of companies are haphazardly successful despite themselves. It is not super clear what they are trying to accomplish or how exactly they are doing it. They have a few tricks that work, that they are decent enough at executing, that they are still around. If you walked in and asked them to describe their strategy, their work streams, their goals, the metrics, and the people doing the work, they would either stare blankly or laugh. It would take them weeks to put together a project to find out. Then months to do the project.

That is the company AI cannot help. Not because the model is weak. Because there is nothing to point at.

A sure sign of a flailing company is when departments spend massive time claiming to have those answers but the answers significantly change every quarter, because everything is constantly in flux. People are writing whatever makes them look good. Weeks of preparation, discarded weeks later, restart.

Compare to the small firm that wins. They can tell you in one breath. The problem they solve for customers. Why existing solutions are weak. The goals they have for the year. The metrics that prove it. The challenges blocking those metrics. The strategy to overcome those challenges. The projects implementing that strategy. The work happening this week. By which person. At what cost. Crucially, they give the same answers across different quarters. The shape is stable. The tactics flex.

This is what I mean by steering ability. Strategy plus monthly input goals plus a cadence to course-correct. Inputs you control. Not output goals like "hit 200K in May" but input goals like "ship 30 prospect emails per week," "publish 4 videos," "run 2 workshops." Outputs are observed. Inputs are committed.

If you cannot run this rhythm yourself, no AI implementation will run it for you. That is why the first weeks of every implementation we run at Strategy Sprints are spent on this clarity, not on tooling. A computer needs direction. So does a five-person team.

Both halves of the popular narrative are wrong. "Wait until AI is mature" is wrong. So is "wait until we are ready." The output gap between firms using this and firms not using this is now visible to your customers, your hires, and your investors. The carry of waiting is no longer zero.

Small is the advantage you have over the conglomerates. You can answer the strategy questions in one room. You can grant yourself permission without a six-month governance review. You can ship a thing this afternoon and watch what happens by Friday.

So ship something small. Today.

One workflow. One agent. One repeatable loop. One decision the AI is allowed to make end-to-end, with you watching, in a sandbox where the worst case is a deleted draft.

The size of the thing is the protection. Small means reversible. Small means you learn before you commit. Small means you earn the right to grant the next permission, on the basis of evidence, instead of a vendor's pitch deck. Small means the loop closes around you, not around the vendor.

The owners who will win the next eighteen months will not be the ones with the biggest AI deployments. They will be the ones who shipped 30 small things, kept the 7 that worked, killed the rest without ego, and built their own steering muscle in the open.

There is one more move that costs nothing and is doing more for the few who use it than any feature in any tool.

Build something. Share it the same day. When someone gives you feedback, answer them in one to five minutes.

Three examples I keep going back to. Boris Cherny built Claude Code first inside Anthropic, then shipped it externally in February 2026, and the team replied to user feedback before the user had finished their next coffee. I built Sprint Club first for our own internal use, then opened it to members in December 2024. Strategy Sprints built the 200K Club in October 2020 the same way: tested it inside the team, then opened it to a small first cohort, then iterated weekly with the people who showed up. Every one of those started small, shipped fast, and moved on what users said inside the same day.

The product compounds in front of the people watching it compound. Trust forms in real time, not in a quarterly newsletter.

Most users have learned that product feedback lands in the void. They submit a form, hit send on an email, post in a community, and never hear back. Or they hear back six weeks later from a support template. So they stop bothering. The feedback loop you needed most quietly dies and you never know why.

You change that by answering inside five minutes the first time, and inside six hours every other time. The first reply is the one that sets the relationship. The person who took ten seconds to write you a sentence now feels seen, heard, and understood. They do something powerful with that feeling. They write you a longer sentence next time. They tell their friend. They start treating your product like theirs.

That is the loop that compounds. Not the model. The loop.

This is the culture worth building in the age of AI. Trust. Being seen. Being understood. Being heard. The conglomerates cannot do this. They have ticket queues, SLAs, tier-one support, and weeks of latency baked into their org charts. You have an inbox and a five-minute window. That is the asset.

So ship it today. Tell three people. When they reply, reply back before you finish your tea.

The number one question is not what AI can do for you. It is whether your company is in the state where AI can help at all. If it is not, that is the work. This week. Something small. Today. Then share it. Then answer in five minutes.

If you want a second pair of eyes on which small thing to ship first, book a 30-min coffee with me: calendly.com/simonseverino/coffee-with-simon. Bring the five questions. Leave with the first thing shipped.