Build your AI-first workflow and pull ahead at work
It’s not about who has access to AI tools. It’s about who’s getting good at using them.
TL;DR
The gap: AI-using workers are 14% more productive, but there’s a gap between occasional and power users. The myth: Success comes down to applied judgement and task selection. The Canadian context: Despite high adoption, many Canadian firms struggle to see ROI due to a lack of structured AI fluency.
A recent study of customer support workers found that employees using AI were about 14% more productive than those who weren’t. While that’s impressive, the workers who used the tools the most kept improving over time, while others levelled off.
By the end of the study, the gap between colleagues doing the same job with access to the same tools had measurably widened. That’s what experts are calling an AI productivity gap.
Beyond access: The compounding returns of AI fluency
For a while, the conversation about AI and work was mostly about access. Once the tools were widely available, the thinking went, those with the tools would benefit equally. But a report from Anthropic that analyzed how people used Claude tells a more complicated story:
Experience matters: Users with 6+ months of experience have a 10% higher success rate than newer users.
Complexity shift: Experienced users move beyond simple queries to high-value, complex problem-solving.
The compound effect: A 10% gain in efficiency daily means more visibility, better projects, and faster career progression.
In other words, someone who can turn around a first draft, a research summary, or a project brief in half the time isn’t just more productive in isolation. She’s more visible, more useful to her manager, and more likely to get tapped for better projects.
Why practice outperforms prompting
There’s a whole genre of content online about how to write the perfect AI prompt, as if there’s a magic phrase that unlocks better results. Research suggests that’s not really where gains come from.
More experienced users aren’t writing fancier prompts. The top performers are developing applied judgement. This involves:
Task selection: Knowing which tasks to use AI for.
Problem framing: Defining the parameters of a business problem.
Output critique: Knowing when to push back on mediocre AI results.
Quality standards: Having a high bar for what good looks like in the final product.
The OECD flagged that early AI adoption is reinforcing skill divides in the labour market. In the current labour market, repetition creates the divide. In other words: practice matters, and people who practice more are pulling ahead.
What this looks like in Canada
Data from KPMG shows that Canadian workers are using generative AI tools more than ever, with many reporting real time savings and productivity gains. But separate KPMG data found that while most Canadian businesses are experimenting with AI, only a small fraction report meaningful returns on those investments. The gap between “we have the tools” and “we’re benefitting from them” is big.
How to build your AI-first workflow
An AI-first workflow isn’t a complex productivity system, it’s a set of habits. It means making deliberate choices about where AI fits into the work you already do.
Ask: Where could AI help me think, structure, or move faster here?
That small moment of intent is what separates occasional use from real leverage.
Here’s how to build it in a way that holds up in your day-to-day.
1. Start with your “anchor tasks”
Most people fail here because they try to overhaul everything at once. Start with your week and identify the tasks that are:
Frequent
Predictable
Slightly annoying to start
The friction test:
If you delay a task because you don’t know how to begin, it’s a strong candidate for AI.
In practice, this usually looks like:
Drafting (emails, briefs, proposals)
Summarizing (meetings, research, long documents)
Structuring (outlines, agendas, project plans)
Pick two 2 or 3. Not 10. The goal isn’t coverage. It’s consistency.
2. Give it enough context, and define what “good” looks like
“Use better prompts” is vague advice. What matters is what you tell the system about:
Who this is for
What you’re trying to achieve
What constraints matter
A simple way to think about it is: What would a new colleague need to do this well?
That includes tone and audience, but also intent. Not just:
“Write this in a professional tone”
But:
“This needs to convince a VP to approve X, with limited time and high scrutiny”
The move from tone to outcome is where the quality changes.
If you have a specific voice, show it an example.
If you have a standard, state it clearly.
The system gives you speed.
You’re still responsible for direction.
3. Treat AI as a collaborator, not a vending machine
The people getting real value aren’t asking once.
They’re iterating.
They’ll say things like:
“This is too formal, make it more direct”
“The structure works, but the opening is weak”
“Cut this by 30% and sharpen the argument”
This isn’t extra work. It is the work. You’re shaping output in real time, the same way you would with a junior team member or a first draft.
Fluency comes from that back-and-forth, not from finding a perfect prompt.
4. Verify before you trust
AI is designed to sound plausible, not to be correct.
If an output includes facts, claims, recommendations, or anything that could have consequences, you need to check it. That might mean:
Scanning for gaps
Cross-checking key details
or sense-checking it against your own experience
An AI-first workflow without a verification step isn’t efficient, it’s risky.
5. Know where it breaks down
Used well, AI can accelerate thinking. Used poorly, it can flatten it. It tends to struggle with:
Navigating internal politics
Making trade-offs with incomplete information
Understanding history, nuance, or stakeholder dynamics
Taking accountability for decisions
In other words: the parts of your job that are actually your job. If you outsource too much here, your work starts to feel generic, and people can tell.
6. Build repetition into your week
This is where your inflection point happens.
Your anchor tasks stop being things you could use AI for and turn into defaults.
Every time you draft an email, summarize a meeting, or plan a project, start with AI.
Over time, you get faster at giving context, spotting a weak output, and steering toward what you need. That’s fluency, and it’s something you build through repetition.
7. Don’t keep it to yourself
One of the fastest ways to level up isn’t more tools. It’s seeing how others use them.
Instead of general tips, what spreads is specificity:
“Here’s how I prep for meetings now”
“Here’s the prompt I use for weekly planning”
“Here’s what worked and what didn’t”
Teams that share these patterns improve faster because they learn from each other in real time.
8. Then, layer in more advanced workflows
Once the basics are in place, you can start to build around them.
For example, using a tool like Microsoft Copilot to generate a daily briefing:
Top priorities
Calendar overview
Inbox triage
Relevant industry updates
This can be useful, but it’s still a starting point. Based on observed patterns, these summaries can miss nuance or mis-rank what actually matters.
Treat them as a draft, not a decision. You’re still the one setting priorities.
The change
An AI-first workflow isn’t about doing less thinking.
It’s about moving the effort away from starting from scratch and toward shaping, evaluating, and deciding.
The people who benefit most aren’t the ones using AI the most.
They’re the ones who know where it helps, know where it doesn’t, and stay accountable for the outcome.
The divide is forming
We’re beyond debating whether AI is changing work - it’s changing everything.
Having access to the tools isn’t the whole advantage. Getting better at using the tools consistently, on real work, over time is what turns access into something that helps you in your career.
I put together a mini workbook that walks through the anchor task audit, the context framework, and a 30-day practice grid. It’s designed to be printed or kept open on your desktop. Pay what you want, including nothing.
AI in the news
Anthropic acknowledges testing new AI model representing ‘step change’ in capabilities, after accidental data leak reveals its existence (Fortune) Anthropic is testing a new, more powerful AI model, reportedly called Claude Mythos or “Capybara,” that represents a major leap in capability, particularly in reasoning, coding, and cybersecurity, but is being released cautiously due to its risks. A data leak revealed internal concerns that the model could significantly accelerate cyberattacks if misused, highlighting a growing tension in AI development between rapid capability gains and the need for stronger safeguards.
Wikipedia bans AI-generated articles (The Verge) Wikipedia has banned editors from writing or rewriting articles with AI, citing repeated violations of its core policies around accuracy, sourcing, and neutrality. AI can still be used in limited ways, like copyediting or translation, but only under strict human oversight. This reflects a broader push to protect information quality as AI-generated content floods the web.
OpenAI is scrapping the Sora app to chase bigger AI goals (Business Insider) OpenAI is shutting down Sora as a consumer app and API, due to unsustainable costs, mounting IP issues, and growing compute demands. The company is reallocating resources toward more strategic priorities, especially AI agents and robotics, signalling a shift away from creative video tools toward systems designed to act in the real world. The decision caused their Disney deal to collapse.
Esther Perel provided couples therapy for a man and his AI ‘girlfriend’ and now I fear for the human race (Guardian) Dramatic headline, yes, but it’s happening: people all over the world are having “romantic” relationships with their AI.
Hey — I came across your writing and really liked how you think.
I’m exploring something similar from a different angle — writing about human behavior through a system design lens (like debugging internal patterns).
Just started publishing on Substack. If you ever get a moment to read, I’d genuinely value your perspective.
Also happy to support your work — feels like there’s an interesting overlap here.