The AI Memos
A look at the "AI or bye bye" re-wiring of tech corporate culture
Want to get better at “AI”? Play with dozens of LLMs and try out different prompts and configurations to see what moves the needle for your goals, all in one convenient location: https://evals.gnodesk.com/. Join the waitlist to be one of the first!
Update 5/29/2025: The AI memos keep coming. The Fiverr CEO (the company that stands to lose the most from AI automating simple white collar work) and the Canadian Prime Minister have sent their missives, respectively. Added below.
While it seems obvious that EVERYONE ELSE should be using AI to do their work, it is hard to swallow the pill yourself, for a number of very ‘human’ reasons — change makes us uncomfortable, change is hard, there is a sunk cost we don’t want to give up on, we are scared of being replaceable, etc. etc. Most of all, large companies operate like pareto-efficient bureaucracies and bureaucracies know only one thing: self-preservation. AI threatens this bureaucracy.
All this to say that there is a silent internal resistance in every company from really adopting AI into their work even as they all build AI products, whether it stems from the above human reasons or from brass-tacks realities of IT provisioning, privacy, legal, budget, etc. friction.
So Tobi (Shopify CEO, also known for his no-nonsense directness and being extremely hands-on) sent this internal memo to counteract all this friction internally. The memo leaked, and Lulu Cheng Meservey (the Silicon Valley comms savant) called it a “courage cascade”, probably nodding to the notion of a “preference cascade”.
Here’s Tobi’s memo: https://x.com/tobi/status/1909251946235437514
Then Luis von Ahn (Duolingo) sent a similar memo: https://www.linkedin.com/feed/update/urn:li:activity:7322560534824865792/
And just today Aaron Levie (Box) sent another one to his employees: https://x.com/levie/status/1917970814441607623
I must have missed this in the original version of this post, but the Fiverr CEO Micha Kaufman sent his memo a day after Tobi. This is particularly fired up because Fiverr stands to lose the most from AI automating away the low-hanging white collar jobs: https://x.com/michakaufman/status/1909610844008161380
Most recently, governments are jumping on the wagon: the Canadian PM sent a letter to his ministry highlighting their priorities. Amongst these, “Government itself must become much more productive by deploying AI at scale…” https://x.com/CanadianPM/status/1925335891146395734
We’ve seen more culture shocks to Silicon Valley tech in half a decade than we have in two. The first 20 years were gradual — recovery from the dotcom bust, then Google leading the way with perks and 20% time, peer bonuses, “hire smart people and get out of their way” approach, and then came the excesses of ZIRP. The last 5 years has seen Brian Armstrong (Coinbase) drive a hammer into the increasing political activism in tech, and Elon break all rules and lay the blueprint of the post-ZIRP culture with his very public “hardcore” breaking-and-remaking of Twitter. And now AI is looking to be the 3rd shock.
I’ll wager this will be the most impactful and most widely-spread. I’m going to keep this post updated with other memos that get sent out / come into the public domain.
Obviously I asked OpenAI’s O3 what it made of these memos (with a few pointed questions), and here is a very very interesting take. Note: what follows below is AI-generated.
What each CEO is telling employees to change
1. Shopify
AI use moves from “nice-to-have tinkering” to a non-negotiable core skill. Every prototype in a “Get-Sh💥-Done (GSD)” cycle must start with AI exploration, and performance reviews will now grade how fluently you prompt, context-load, and share AI learnings. Before asking for extra headcount, teams must show why an autonomous agent can’t do the work. In short: climb the AI curve or slide out of relevance.
2. Box
Each org must publish an explicit AI strategy and bake it into Q-level OKRs. Use sanctioned tools to strip drudgery, speed decisions, and free budget—then re-invest those savings in higher-impact projects (“keep what you kill”). Expect constant internal experiments, mandatory “AI up-skilling,” and rigorous security / human-in-the-loop governance for any proprietary data. The metric isn’t lower cost; it’s more strategic work shipped faster.
3. Duolingo
Rewrite workflows so AI, not contractors, handles repeatable content and feature creation. AI proficiency becomes a hiring, performance-review, and headcount gate; teams get slots only when they can’t automate further. Functions are tasked with ground-up redesigns (willing to take small quality hits) to unlock scale—e.g., generating lessons and AI-tutored video calls. Goal: free humans for creative pedagogy while AI removes bottlenecks.
4. Duolingo
Radical wake-up: “AI is coming for your job.” Master field-specific AI tools, become a prompt engineer, and deliver more output at higher quality—or face an imminent career pivot. Use AI to make the org more efficient before any new hiring and pitch ideas proactively; nobody will schedule a meeting just to hear you out.
What’s uniquely different about each playbook
1. Shopify
Merchant-centric multiplier
Treats AI as a force-multiplier for its “10×” builders and, by extension, for entrepreneurs on the platform. The litmus test is: could an autonomous agent already be on this team?—a framing geared to shipping commerce tools faster than competitors.
2. Box
Governance-backed reinvestment loop
Leans hardest on security, privacy, and sanctioned tooling—logical for an enterprise-content company—and pairs every efficiency win with a “keep what you kill” reinvest policy. That marries CIO-grade trust with a startup-style permission to re-deploy freed budget.
3. Duolingo
Content-scale accelerator
Uses AI to explode the volume and variety of pedagogical content (the company’s real moat) and to pioneer tutor-level features once considered impossible. Distinct moves: phasing out external contractors and accepting minor quality trade-offs for first-mover scale.
4. Fiverr
Survival-of-the-sharpest credo
Leans hardest on personal fear-of-obsolescence: master AI or be “doomed.” Emphasizes individual up-skilling across all roles and a no-hand-holding culture of proactive idea pitching.
Cultural levers the memos are really pulling
(I asked O3 to break down the memos based on the cultural DNA of each company)
1. Shopify
Primary lever: Fear-of-stagnation → Growth-mindset status game
Why it resonates internally:
Tobi frames AI as the only way to stay in Shopify’s Darwinian “red-queen race.” Opting out isn’t neutral—it’s “slow-motion failure.” That taps high performers’ fear of falling behind peers and missing the next promotion cycle.
Examples from the memo:
“If you’re not climbing, you’re sliding.”
AI usage questions baked into performance reviews.
Teams must prove AI can’t do the work before asking for headcount.
2. Box
Primary lever: Cost-to-value trade-off → Autonomy with guardrails
Why it resonates internally:
Aaron leverages Boxers’ desire to kill bureaucratic toil and keep the spoils. The “keep what you kill” rule converts efficiency wins into discretionary budget, granting teams agency. Pairing that with strict data-governance talk reassures risk-averse enterprise buyers and employees.
Examples from the memo":
Reinvest AI savings in “more strategic things.”
“Sanctioned tools only” and human-in-the-loop for proprietary data.
Org-level AI OKRs = public accountability.
3. Duolingo
Primary lever: Mission-speed urgency → Builder’s pride
Why it resonates internally:
Luis ties AI directly to Duolingo’s founding story (the 2012 mobile bet) and to its mission of free education. The lever is pride in moving the learning curve for millions faster than anyone else—even if that means “small hits on quality.”
Examples from the memo:
Replacing manual content creation to “owe it to learners.”
Contractor sunset and performance gates pivot talent toward creative work.
“Betting on mobile made all the difference… AI is the same.”
4. Fiverr
Primary lever: Fear-of-obsolescence realism
Why it resonates internally:
Radical candor fits gig-economy hustle DNA.
Examples from the memo:
AI is coming for your job… be exceptional or pivot.
Second-order effects to watch
Talent split: Fiverr’s fear-based memo may push slower learners out faster than Shopify’s status framing.
Shadow tools risk: Box’s sanctioned-only stance vs. Fiverr’s “try multiple solutions” could create hidden stacks in both directions.
Quality vs. brand: Duolingo’s quality trade-offs and Fiverr’s pressure cooker both risk user trust if AI output slips.
(Back to Umang-generated content)
This is just the tip of the iceberg. It took years for companies to realize that modern smartphones were not that different from company-issued laptops and for work to move from the laptop to the mobile. IT provisioning, having new mobile-first work apps, new work habits (always on, more informal and terse interaction, etc) were all hurdles we had to cross. Using AI will be similar. 2025 will be the year of experimentation and 2026 is when we’ll see the AI-native corporate operating system start to solidify. Year-boundaries are convenient to visualize but arbitrary — I bet the change will happen faster.