On April 7, 2025, an internal memo from Shopify CEO Tobi Lütke surfaced publicly after he reposted it himself, and one line did the rounds in every founder group chat by morning: reflexive AI usage is now a baseline expectation at the company. The memo went further than a vague encouragement. It said teams must demonstrate why a task could not be done by AI before they would be granted additional headcount or resources, and that AI fluency would become part of performance and peer reviews. Whatever you think of the framing, the signal was unmistakable. A founder-led, profitable, multi-thousand-person public company had just made AI competence a condition of employment rather than a nice-to-have. As the founder of Softechinfra, I run a services business where margins live and die on how efficiently a small team ships, so I read that memo less as a headline and more as a question pointed directly at every SMB owner: if a company that size can treat AI as default, what is your excuse, and what is your plan? This guide is the answer I gave my own team—what "AI as baseline" actually means once you strip the drama, and a 90-day plan to get there without burning trust.
What "Reflexive AI" Actually Means
The phrase that travelled was "reflexive AI usage," and it is worth slowing down on, because most teams hear it and imagine either nothing changing or everyone being replaced. Both readings are wrong.
Reflexive means automatic, the way an experienced developer reaches for version control or a marketer reaches for a spreadsheet. It is not a heroic initiative with a steering committee. It is the quiet default that, before drafting the first version of a job description, a contract clause, a support reply, a SQL query, or a campaign brief, someone asks the model first and then edits. The human still owns the judgment, the accountability, and the final word. The model owns the blank page.
That distinction matters because the failure mode of "everyone must use AI" is people pasting raw output into client-facing work and shipping confident nonsense. The Shopify framing—AI to multiply a competent person, not to replace the competence—is the version that survives contact with real work.
Why This Matters More for SMBs Than Enterprises
It is tempting to file a memo from a company Shopify's size under "big-company problems." The opposite is true. The smaller you are, the more an AI baseline moves the needle—and the easier it is to actually change behavior.
A large enterprise changing a default fights procurement, legal review, works councils, and the inertia of thousands of habits. A ten-person firm changes a default in a Monday standup. You have no legacy AI strategy to unwind, no committee to convince, and your people already wear multiple hats—exactly the profile AI assists best, because the bottleneck is almost never headcount, it is the sheer breadth of tasks each person carries. The same dynamic plays out at the level of the whole organization, which is why a structured rollout matters; our enterprise AI transformation framework covers the governance side that even a small team eventually needs.
The Honest Risk List (Read This Before the Plan)
A mandate without guardrails manufactures exactly the disaster that makes leadership distrust AI for a year. Before any rollout, put the failure modes on the table where everyone can see them.
- Confidentiality. Decide what may and may not be pasted into which tools. Client data, credentials, and unreleased plans need a clear rule, ideally enforced by using business-tier tools with data-retention controls rather than consumer accounts.
- Hallucination. Anything factual—numbers, citations, legal or medical claims—must be verified by a human before it leaves the building. "The model said so" is never a source.
- Skill atrophy. Juniors who never write the first draft never learn to write. Use AI to review and stretch junior work, not to replace the practice that builds judgment.
- Over-automation. Some tasks are cheaper done by hand than scripted, prompted, and maintained. Automating a five-minute monthly task is a hobby, not an ROI.
- Disclosure and trust. Agree where AI assistance is fine silently (internal drafts) and where it must be disclosed or kept human (sensitive client comms, anything regulated).
A Realistic 90-Day Adoption Plan
Skip the moonshot. The goal of the first quarter is not transformation; it is to make "ask the model first" a boring, trusted habit on real work, with the guardrails above already in place. Three thirty-day phases.
Days 1–30 — Map and equip
Have each person list the five most repetitive or blank-page tasks in their week. Cluster them—you will find drafting, summarizing, data wrangling, and code scaffolding everywhere. Pick business-tier tools with proper data controls, write your one-page acceptable-use rules, and name one champion (not the most senior person—the most curious one). No mandate yet. The deliverable is a shared task inventory and a safe sandbox.
Days 31–60 — Pilot on real work
Each person picks two tasks from their list and does them AI-first for a month, keeping a rough log of time saved and edits required. Run a weekly fifteen-minute show-and-tell where people share prompts that worked and outputs that embarrassed them. Build a shared prompt library from the wins. This is where reflexive usage is actually born—through repetition on work that matters, not a training video.
Days 61–90 — Standardize and decide what to build
Promote the proven workflows into team defaults and bake them into your checklists and onboarding. Now look at the tasks where copy-paste prompting is too slow or error-prone and decide which deserve real automation—an integration, an internal tool, a pipeline. That is the line where casual usage ends and engineering begins.
The handoff at day 90 is the important one. Reflexive AI usage gets you efficiency at the level of the individual. The compounding gains come when you encode the winning workflows into systems your whole team uses by default—which is exactly the kind of work our AI automation services exist to do. We treat the prompt library and the task inventory from your pilot as the spec.
Casual Use vs. Built Automation: Where to Draw the Line
Most teams blur these two and get frustrated, either babysitting a chatbot through a task that should have been a button, or building a fragile script for something they do twice a year. The decision is not about how impressive the AI is; it is about frequency, stakes, and stability.
| Signal | Keep it casual (prompt by hand) | Build real automation |
|---|---|---|
| Frequency | Occasional, varied, one-off | Daily or many times a week |
| Inputs | Different each time, need judgment | Predictable, structured, repeatable |
| Stakes if wrong | Low—a human reviews anyway | High or invisible—needs guardrails in code |
| Volume | A handful of items | Hundreds, beyond manual capacity |
| Maintenance cost | Higher than the task is worth | Easily repaid by time saved |
A practical example from how we work internally: routing and summarizing the steady stream of internal updates, requests, and documents across a growing team is high-frequency, structured, and high-volume—a textbook automation candidate, and the kind of workflow our Intranet product was built to absorb so people stop doing it by hand. By contrast, writing a nuanced reply to a single unhappy client is low-frequency and high-judgment: prompt for a first draft if it helps, then write it yourself. Knowing which is which is most of the skill.
How to Measure It Without Theater
Counting "AI logins" or "prompts sent" rewards activity, not outcomes, and your sharpest people will resent being measured by it. Tie measurement to work that already mattered.
- Cycle time on standard deliverables. Is a first draft of a proposal, a support reply, or a feature spec landing faster than it did in January? That is the real signal.
- Backlog burn. Are you clearing tasks that used to sit untouched because nobody had the hours? AI-as-baseline shows up as ambition you can finally afford.
- Quality, not just speed. Track rework and client revisions. If speed went up but rework did too, your guardrails are too loose.
- Qualitative pulse. Ask the team monthly which tasks AI genuinely improved and which it made worse. Retire the bad fits without shame.
Our CTO Hrishikesh Baidya applies the same eval discipline here that we use when adopting any new model: define the task, baseline the current cost, then measure the change on that specific task rather than chasing a generic benchmark. The principle is the same whether you are evaluating a tool or a habit, and it is the throughline of our broader guide to building AI features in production—measure on your work, not the leaderboard.
The Durable Takeaway
The specific tools named in any 2025 memo will age fast; the models, the apps, and the price points are already moving. What will not age is the underlying shift the Shopify note made visible: AI fluency is becoming a baseline professional skill the way spreadsheets and search once did, and the organizations that treat it as a deliberate, guard-railed default will out-execute the ones still debating whether it is allowed. For a small team that is not a threat—it is the best leverage you will get this decade. Write your one-page rules this week, run the task inventory next, and let the 90-day plan turn a memo you read about into a habit you actually have. For the wider quarter that produced this moment, our Q1 2025 AI recap for business sets the scene.
Want AI as a Real Default, Not a Slogan?
We help SMB teams turn an AI mandate into working systems—acceptable-use rules, a 90-day adoption plan, and the automation that turns winning prompts into tools your whole team uses.
Talk to Our AI Automation Team →
