The term “AI-first” has been everywhere for a few weeks now. On LinkedIn, in vendor pitches, in business articles, from consultants proposing full overhauls billed at full-overhaul prices. And from leaders who are asking, rightly, what it actually means for their SMB or organization.
I want to bring some order to this. Not to have the last word on a buzzword, but because the current confusion is costing time and money to leaders who genuinely want to get this right.
This article builds on what I shared in L’Architecte, my newsletter, in June 2026. It opens the subject further, lays out the three competing readings of the term, and offers the translation I find to be the only truly useful one for the leaders I work with: AI as a first reflex.
The same word, sold three different ways
When a vendor, an article or an employee talks to you about AI-first, they mean one of three things, and the price tag varies by a factor of one hundred depending on which one.
Reading 1: AI-first as a marketing posture
A banner on the website. A press release. A line in the investor or client pitch. “We are an AI-first organization.” Nothing changes inside. People work the way they always have. Processes stay the same. So do the tools. It is cosmetic, and it stays harmless until someone believes it.
The risk is that it creates an expectation nothing supports. The client who calls you because you are AI-first will be disappointed once they see your real day-to-day. The employee who joined for that reason will feel misled after three months. And the leader who ends up believing it stops seeing the gap between the banner and reality.
Reading 2: AI-first as an organizational overhaul
The opposite of the first. A major program lasting 12 to 24 months. New architecture, redefined roles, a six-figure platform, a governance committee, consultants on constant rotation. It is expensive, it is slow, and it is often disappointing, because the target moves faster than the project.
When you see an AI-first proposal that opens with an overhaul of that size, put your pen down. Not because an overhaul is bad in itself (sometimes it is even justified down the road), but because it should never be the first step for an SMB or organization that does not have the reflex in place yet.
Reading 3: AI-first as a decision reflex
The only reading that is both useful and within reach for most leaders. Becoming AI-first means building a simple discipline into your team’s culture: before attacking a task, you ask whether AI can give you a hand, and how.
It is the equivalent of buckling your seatbelt. Nobody thinks about it, you just do it. Nobody needs to rebuild the car to adopt the reflex. You train it, you keep it top of mind, you make it automatic, and it becomes part of how the team works.
That third reading is the one I use here.
The translation I actually use: AI as a first reflex
When a leader asks me what AI-first means, I translate it as AI as a first reflex. Not a first project. Not a first budget line. A first reflex.
This translation clears up several ambiguities at once. It rules out the overhaul (a reflex is built, not ordered). It rules out the cosmetic (a reflex shows in action, not on a banner). It rules out replacement (the reflex includes the clear-headed decision not to use AI when it is not appropriate).
Above all, it opens the door to immediate implementation. You do not need to wait for a budget or a platform to start training this reflex in your team this week.
What AI-first is not: three distinctions
Before I explain how to build the reflex, I want to close three doors that stay open too often.
AI-first is not AI-only
No serious person is proposing to replace your teams, your proven processes and your human decisions with AI. If you see someone selling that, you know what to do (put your pen down). The AI-first reflex includes the clear-headedness to keep people up front when that is what best serves your operations, your client relationships and your quality.
AI-first is not AI-right-now
You do not have to deploy 14 tools this quarter. The reflex is built over time. A team that tries to adopt five new platforms in six weeks burns out, gives up, and keeps a bad memory that closes the door for two years. Better to take one business function, one tool, six weeks of observation, then expand.
AI-first is not an overhaul
You do not have to shut down operations, redraw your org chart, replace your tech stack and spend six months in committee before you begin. The reflex is built in the real tasks of real teams, not in a boardroom.
How the decision reflex works
Concretely, what does the reflex look like once a team has it?
Before an employee starts a task that matters (writing a report, prepping a meeting, sorting a pile of email, structuring a tender, analyzing a client file, drafting a proposal), they take thirty seconds to ask four questions.
- Can an AI tool give me a hand here? Not necessarily do the task for me. Build an outline, rough out a draft, summarize a stack of material, compare two versions, check my logic.
- If yes, which one? Not just any tool. The one approved here for this kind of data. The one where I can enter this information without creating a problem with confidentiality, Quebec’s Law 25 or intellectual property.
- If no, why not? Because the data is sensitive, because the call is too fine, because a human still does it better. A clear no is a full answer, not a failure.
- What do I note for next time? A small trace, even a mental one. “Next time I do this kind of task, I know what works and I know what jams.”
Four questions. Thirty seconds. That is the reflex.
At first, the team forgets one time in two. After four weeks, one time in ten. After six months, it is automatic, and productivity, quality and calm have risen together.
Governance and security live inside the reflex
This is probably the most important point in this piece. When AI-first comes up, many leaders think governance, information security, Law 25 and data classification are a separate file, to be handled in parallel, by someone else, later.
No.
These questions live inside the reflex itself. The second question above (“which tool is approved for this kind of data?”) is meaningless unless you have already set a minimal framework: what counts as sensitive data in your organization, which tools it can go into, and who decides.
What you need is not an $80,000 platform. It is:
- A short list of approved AI tools for your organization, with their confidentiality level.
- A one-page usage guide that says what can go out and what must stay in-house.
- A clear point of contact on the team when someone is unsure (“I have a client file, can I run it through this summarizer?”).
- A light quarterly review to adjust the list as new tools come out.
That is what building governance into the reflex looks like. It cuts across everything, it is built in, and you cannot outsource 100% of it to a consultant who drops by once a year.
What I am learning by building my own
I am going to tell you what I am living through in my own shop right now, because it is the most honest example I can give.
I am currently building an internal tool for my own operations. A layer on top of Microsoft 365, to force myself to apply this AI-first reflex in my daily work as the head of a small shop. I am not announcing it, not selling it, not naming it here. I am building it for me, to honestly test what AI-first means when you are the one in the seat.
Four things I am finding along the way, shared as they are.
First finding: the reflex does not arrive by magic
Even when you are convinced, even when you work in the field, you literally have to force yourself for the first few weeks. I still regularly catch myself opening a blank Word document and starting to write without asking my AI assistant what it thinks. The old reflex (doing it by hand, start to finish) is so deeply rooted that the new one needs time to settle in.
What this means for you: do not expect your teams to adopt the reflex in two weeks because you announced it at a meeting. Plan for reminders, support and patience.
Second finding: it is not the tool that unlocks things, it is the framework
My internal tool does simple things. It helps me sort what deserves my attention, prepare my answers, structure my notes. None of those functions is revolutionary taken one by one. What changes is that the framework is set, the good reflexes are guided, and sensitive data stays where it belongs.
What this means for you: do not look for the miracle tool. Look for the minimal framework that makes the good reflex easy and the bad reflex hard. A one-page guide, three approved tools, one point of contact. You can start this week.
Third finding: governance is part of the build, not beside it
Every function I add to my internal tool forces me to clarify where the data lives, who can access it, what may go out to an external model and what must stay in-house. Law 25, information classification, permission management, prompt traceability. Those are architect questions, and they rise at the same time as the reflex, not after.
What this means for you: if you start your AI-first reflex without these questions in mind, you are building a debt you will pay back with interest. Better to start small, clean and structured than ambitious and fragile.
Fourth finding: it is an ongoing build, not a project
There is no end date. There is a cadence of improvement. Every week, I watch what works in my reflex and what jams, and I adjust. My internal tool evolves with my own adoption. It is exactly the stance I recommend to the SMB and organization leaders I work with.
What this means for you: do not ask your management committee for a delivery date to “be AI-first.” Ask your team for a cadence of observation and adjustment instead, and commit to keeping it up.
What to remember
Three ideas to hold on to.
AI-first means AI as a first reflex, not AI everywhere right now. The reflex includes the clear-headedness to say no to AI when it is not appropriate. An AI-first team is a team that asks the question, not a team that always answers yes.
The reflex is built in daily work, not in a big project. Six weeks of quiet discipline on one business function beat eighteen months of overhaul. Start small, watch, adjust, expand.
Governance, security and Law 25 are part of the reflex, not beside it. Going AI-first with no framework on what you send, where, and under which rules is simply careless. The framework is minimal (one page is enough to start), but it is non-negotiable.
Available resources
A resources page on my site brings together the tools I share to help leaders start structuring their digital operations and their use of AI. Available with no email required:
- One-page Starter Pack, a minimal AI governance framework in seven questions, to fill out with your team in one hour. An excellent starting point for anchoring the AI-first reflex.
- 18-page Practical Guide, the in-depth version for organizations that want to go further on structure and compliance.
Everything is in one place on the resources page: stevejohnston.co/en/resources/.
Let’s talk
If you want to talk it through, look together at what the AI-first reflex would look like in your SMB or organization, or simply validate your minimal governance framework, I would be glad to do it with you: stevejohnston.co/en/contact-us/
