What do Shopify, Duolingo, Fiverr, and Google have in common?
They’re all eager to be thought of as AI-First companies. Shopify, Duolingo, and Fiverr are the latest crop, all hitting the news recently by issuing company-wide memos requiring everyone (EVERYONE!!!) to use AI.
Where does Google come in? They’re the OG of AI-Firstitude.
Nine years ago at Google, I played a key role in the world's first AI-First company’s successful AI-First transformation. (What a sentence!) I mention this here because AI-First isn't just a buzzword for me. I've lived it from the beginning and so of course, Kozyr, the company I founded after Google, has been AI-First from Day 0.
But, unlike these cool cats Shopify and Duolingo, we've never had an AI-First company-wide memo at Kozyr. Last week I fixed that!
If you’re inspired to skip reading the rest and fire off your own company-wide memo, I encourage you to invite some nuance about what "AI-First" can mean for you and your team. For example, Shopify's memo and their approach to an AI-First culture has plenty to admire but also plenty to be wary of (read this).
I offer you the memo I wrote to my staff in the hope that you’ll find it a sober and practical counterpoint to much of the hype du jour.
Here’s the Kozyr Memo:
Please cheat at your job!
At Kozyr, we are AI-First. That means everyone here, regardless of role. So, what is AI-First and what does the company expect of you?
To us, AI-First is an attitude. Or, rather, two attitudes. We want you to embrace them both.
1) “How could I do this better?”
We are tool-using apes and proud of it. There’s no honor in doing tedious things if there’s a safe and effective way to cut out the drudgery or improve the way we do things by adopting tools. Humanity’s history is a history of tool making and tool use. No matter what your role is and no matter what your background is, you can always improve whatever you’re doing. How? Get advice and then follow it. Wisely.
Success is all about priorities. The hardest part of self-leadership is balancing your time between trying new things (which don’t always work, even with AI… and sometimes especially with AI) and delivering your deliverables. Here’s a habit that will take you far: as soon as you’re faced with a high-value (important to your role) or high-cost (repetitive or expensive) task, get a second opinion before you dive in.
If you’re an experienced professional, you’ll have developed the habit of asking a human. Good. Keep doing that. AI is not a replacement for human wisdom.
But with AI, getting advice has never been cheaper.
That’s why you have no excuse not to seek a second opinion on the important stuff. I encourage you to get it from AI.
Here’s how: in your Kozyr Workspace in ChatGPT with Temporary Chat enabled (if you’re new to this, ask your manager), describe (with the confidential details scrubbed out) what you’re about to do and ask for detailed suggestions for improving your approach. Try phrasing your question a few different ways and see what kind of ideas the AI output sparks.
To avoid getting fixated on one answer, ask for 30* suggestions instead of just one and then dig deeper on a few of them. The more thoughtful your questions, the more useful your answers will be. Just like getting advice from people!
So what’s the new hard part of everyone’s job? Knowing what’s worth doing better (and doing at all). Prioritization and clear communication. That’s how you make sure that what you get out of your AI tools is maximally useful to you.
Using a generative AI tool is very easy. If you know how to send a text message, you know how to use ChatGPT. Using it is easy… but using it well takes smarts and experience. The best time to start getting that experience was in 2022. The second best time to start is now.
Some of the time, the suggestions for improvement won’t tell you to use more AI. That’s okay. Using AI where it’s counterproductive is foolish and it’s not what we mean by AI-First. But a lot of the time, you’ll be amazed by ways you can improve your work with AI. In which case… improve your work. Tool use is awesome!
Get in the habit of asking AI these questions about your priority tasks:
How could I do […] better?
How could I make […] easier?
Are there any tools I should know about to help me with […]?
What are 30 ways I might improve […]?
Might [this impossible thing] be possible?
Start by figuring out what’s worth doing, then ask an AI system like ChatGPT how to do it better. Then keep asking smart questions – you’ll be amazed how much time and toil you’ll be able to save yourself and your team.
In many ways, this first kind of AI-First is an extension of a principle developers live by: DRY.
“Don’t. Repeat. Yourself.”
If you find yourself doing something repetitive, that’s a sign you should be asking for advice on how to minimize the drudgery.
To live by this principle, you used to have to learn a programming language first, so DRY was for those who had the patience to learn something like Python, C++, etc. Today, it’s for everyone. Why? Because today’s AI revolution is a language and interface revolution. The hard part isn’t figuring out how to express what you need in an unnatural language. It’s figuring out how to express what you need in your mother tongue. You already know how to speak that, so speak it. Every time you get stuck, ask questions until you get unstuck. Ask about new approaches, new tools, etc… and then take the good advice.
But don’t take the bad advice. AI isn’t intelligent and it does make mistakes, so you are expected to bring the intelligence. Don’t be gullible. Keep your wits about you and don’t believe everything that comes out of an AI system. Ultimately, you’re responsible for how you use your tools.
So that’s the first AI-First attitude: don’t treat your work like a closed-book exam. If asking for advice is cheating, then I’m imploring you to cheat. Figure out how to cheat at your job and then teach all of your colleagues how to do it too. The point of work isn’t suffering, it’s making progress. So go make progress!
Before we look at the second attitude of AI-First, some warnings.
WARNING: AI-FIRST DOES NOT MEAN…
A) AI-First does not mean using AI output without checking it first (and perhaps modifying it if needed).
Don’t assume AI is intelligent. AI is just a tool, it’s not your boss. You bring the intelligence. So don’t follow AI’s advice blindly. Sometimes there’s no good advice for your situation, but it’s always worth asking. Especially when asking has never been cheaper. You’ve got nothing to lose… as long as you fact-check anything that you’re tempted to act on.
Do not fall asleep at the wheel. You're responsible for the quality of your work, with or without AI. AI cannot take responsibility. Ever. You are responsible for what you do with your tools.
B) AI-First does not mean putting sensitive information into external AI systems.
We take this very seriously at Kozyr. As part of working here, you will have been briefed on sensitive information and we expect you to revisit your briefing notes frequently. If you are unsure, scrub out the sensitive info and/or ask a senior member of staff whether your situation falls in this category.
C) AI-First does not mean installing dodgy plugins and programs from disreputable sources.
This goes for both AI and traditional systems. As a professional in this millennium, you should know better than to trust everything you find on the internet. If in doubt, ask the AI Resources team to advise you. For their convenience, remove the confidential details and ask your question of ChatGPT first, then include the answer in your description email for them so they don’t need to start from scratch. (And why not use ChatGPT to generate a first draft of that whole email for you while you’re at it?)
D) AI-First does not mean AI-for-AI’s-sake.
If something isn’t worth doing, then it’s still not worth doing with AI. Never lose sight of the value of what you do and strive to keep that value positive. Use judgment to pick the most fruitful direction, then use AI to get there faster.
E) AI-First does not mean robots.
What’s frowned upon:
i) Not mentioning a tool just because it seems expensive.
Even if it’s expensive today, useful tools will only get better and cheaper tomorrow. Put it on our radar even if you think it’s a bit too pricey. Whenever you discover an awesome new tool, shoot a short note to the AI Resources team so we can add it to our internal solutions list and share the knowledge internally. The AI Resources team will do the heavy lifting of documenting the tool so you won’t need to. Unless you’re part of that team, in which case you’ll be using AI to help you with the boring bits of tool documentation.
ii) Using AI tools and pretending you didn’t.
Your job is safe from self-automation, since there will always be plenty to do for an innovator like you. The change of pace is accelerating, and so are we. The most important things you bring to the table at Kozyr are:
your adaptability and motivation to learn new things (which can’t be automated away)
your experience (which only increases with time spent at the company)
your personality and contribution to our culture (be a joy to work with)
our trust in you (which takes time to earn and is irreplaceable). Since trust will be the grand challenge of tomorrow’s world, we would be fools to undervalue it. We’re not fools.
Earn our trust and there’s no such thing as automating yourself out of a job, there’s only growth. So don’t be a “secret cyborg” who robs us of the opportunity to learn from you. Trust us to be fair and reward you appropriately for your ingenuity. We would love the opportunity to celebrate your clever solution with you.
Now that we’ve gotten the warnings out of the way, let’s continue.
The first flavor of AI-First is for everyone at the company. All of you should be embracing tool use as a way of being better at your work.
This next flavor of AI-First is for automators. Those of you whose work involves setting things in motion that doesn’t require human hands at every step of the journey.
2) “Can we scale the impossible?”
At Kozyr, we are very serious about the quantum leap in responsibility from actions that only affect the individual who’s doing them and actions that reach beyond just you. If there’s truly no way anyone but you (or your consenting team) will be impacted by your mistakes, knock yourself out - you’re an adult. But anything you do where others will join you in bearing the consequences of your mistakes should be approached with the appropriate gravity. Ignorance is no excuse - with great scale comes great responsibility.
Which is why our attitude towards AI-First shifts when we move from individual tool use to automation. In this case, being AI-First doesn’t mean shoving AI everywhere and hoping for the best. It means rethinking what’s impossible. Responsibly.
When, not if.
AI systems are complex. When a complex system touches the real world, expect mistakes. There’s no magic that changes this. That’s why automating with AI should be a balancing act that involves:
Picking applications where mistakes have minimal potential to cause harm OR where all mistakes can be remedied OR where mistakes only affect Kozyr while being profitable on balance OR where the benefit to society eclipses the cost of any mistakes. Judgment calls are required here, which is why AI automation is always at least as much a leadership concern as an engineering concern.
Creating guardrails, safety nets, and control structures that mitigate the risks of mistakes.
Taking responsibility for the downside, not just the upside, of your actions.
Because complexity invites risk, traditional programming is the safer way to approach automation. AI-First doesn’t mean doing things the dangerous way just because we can.
Instead, being AI-First means that before we declare something to be impossible, we first ask if AI puts it on the table for automation.
Being AI-First means revisiting yesterday’s impossible dreams. Instead of saying we can’t do something, at Kozyr, we instead ask:
“What would it take?”
All things equal, we still prefer non-AI solutions when they exist because we love reliability, privacy, and security. Complexity (and thus AI) makes all of these harder to achieve. But the set of problems we can solve without AI is a tiny drop relative to the vast ocean of opportunities that AI puts on the table.
As soon as we reach past the low hanging fruit and take on hard problems with complex solutions, we’ll need AI. Not because we like shiny objects but because we have problems that are important enough to be worth solving and AI is the only way we can solve them.
That’s why I don’t need to ask you to be AI-First. It’ll happen as a natural side effect of thinking bigger. So that’s what I’ll ask of you instead: think bigger! The rest will follow.
Summary
Tool-use: There’s no such thing as cheating at your job. We celebrate tool use! Before diving into an important task, first use AI to get advice on how to approach it better. (But ask smart humans too.) Then follow that advice. Wisely.
Automation: With great complexity comes great responsibility. At scale, complex solutions (like AI) are for things you can’t solve the simple way (without AI). But only if they’re worth solving in the first place. Luckily, that’s a huge category, so instead of assuming something is impossible, first ask if AI might make it possible.
Please steal this memo
If you found my take sensible and you’re itching for something to share in a company-wide blast, you have my permission to help yourself to the Kozyr Memo.
And if you’ve found yourself agreeing with me, please share this with someone who needs to hear it. Or your boss. Or your boss’s boss. Or…
Agentic AI for Leaders
Next week I’ll be teaching live sessions of the Agentic AI for Leaders course along with the authors of the Agentic Artificial Intelligence book. I’ll be doing the kickoff, laying the foundations so that you find the topic approachable and amuseful** no matter what your background is
Enroll here: bit.ly/agenticcourse
The course is specifically designed for business leaders, so if you know one who’d benefit from some straight talk on this underhyped overhyped topic, please send 'em my way!
MakeCassieTalk.com
Yup, that’s the URL for my professional speaking. Couldn’t resist. 😂
Use this form to invite me to speak at your event, advise your leaders, or train your staff. Got AI mandates and not sure what to do about them? Let me help. I’ve been helping companies go AI-First for a long time, starting with Google in 2016. If your company wants the very best, invite me to visit you in person.
Here’s where you’ll find me this month:
May 7 - 8: Orlando
May 9 - 16: New York
May 20 - June 3: San Francisco
June 3 - 7: Santa Fe
June 10 - 14: London
Footnotes
*Why 30? You’ll understand after you try it a few times. At Kozyr, we call this technique “30 bullets” but sometimes the number isn’t 30. Sometimes it’s smaller, sometimes it’s bigger. If you haven’t figured out why this is worth doing after a few days of doing it, ask. Ask a human and, of course, ask ChatGPT too. That’s your new habit, after all. And if it’s very important, get third and fourth opinions with Claude and Gemini, which you also have access to.
**Amusing and useful—the standard I always strive for.