#31 AI-First Businesses: When AI Becomes the Default
Plus: AI Writing Strategies - Improving Your Process and Influencing the Models, and notable developments
Dear Readers,
In this edition, I cover:
Notable Developments
OpenAI dropped plan to convert to for-profit, acquired Windsurf, and hired Simo to lead applications
Google reported good earnings but worries on AI’s threat continue
AI overtakes cybersecurity as top budget priority for global IT leaders
AI platform integrations pose cybersecurity risks
Sequoia’s AI Ascent 2025 conference takeaways
AI-First Businesses: When AI Becomes the Default
AI Writing Strategies: Improving Your Process and Influencing the Models
Enjoy.
OpenAI dropped plan to convert to for-profit, acquired Windsurf, and hired Simo to lead applications
After months of debate and external scrutiny, OpenAI announced it will retain nonprofit control over its operations. Instead of spinning out its for-profit arm as an independent entity, OpenAI’s for-profit division will become a Public Benefit Corporation (PBC) but remain under the nonprofit’s oversight. This compromise preserves the nonprofit’s mission-driven governance while allowing the company to raise capital more flexibly, addressing concerns from regulators, investors, and critics who feared a drift toward pure profit maximization.
Additionally, OpenAI agreed to acquire Windsurf (formerly Codeium) for about $3 billion, its largest acquisition to date. Windsurf is a leading AI-powered coding assistant, known for its fast, context-aware, and modular tooling for software developers.
OpenAI has hired Fidji Simo, CEO of Instacart, as its new CEO of Applications. Simo, who has served on OpenAI’s board since last year, will oversee the newly consolidated Applications division, which brings together the company’s business and operational teams responsible for turning research into products and services used worldwide.
Google reported good earnings but worries on AI’s threat continue
Alphabet, Google’s parent company, reported strong first-quarter 2025 earnings, with growth was broad-based. AI-powered products-like Gemini 2.5 and AI Overviews-drove user engagement and subscriptions, now surpassing 270 million paid users. The company revealed similar monetization rates between traditional search keywords and AI overviews, which was a positive surprise for investors. However, these good earnings were overshadowed by ongoing federal antitrust cases and news that Google searches on Apple's Safari web browser fell over the last two months, the first decline in 20 years. This revives the argument that traditional search is losing market share to AI tools such as ChatGPT and Perplexity.
AI overtakes cybersecurity as top budget priority for global IT leaders
The AWS Generative AI Adoption Index surveyed over 3,700 IT leaders in nine countries and here are two interesting takeaways among many:
More organizations plan to prioritize gen AI spending over traditional IT investments like security.
Financial services are increasingly adopting a hybrid AI solution: either build custom applications on pre-existing models or develop applications on fine-tuned models. Only 25% plan to develop in-house solutions from scratch.
AI platform integrations pose cybersecurity risks
Cybersecurity company Rubrik CEO Bipul Sinha warns that AI integrations introduce new and complex risks. As organizations weave AI into their digital infrastructure, the attack surface expands, giving cybercriminals more opportunities to exploit vulnerabilities-especially through insecure APIs, adversarial attacks, and data privacy lapses. Sinha notes that AI’s power to automate threat detection and response is a double-edged sword: it strengthens defenses but can also be weaponized by attackers to create more sophisticated threats. He stresses the need for robust, AI-specific security frameworks, continuous monitoring, and human oversight to validate AI-driven decisions.
Sequoia’s AI Ascent 2025 conference takeaways
Sequoia’s AI Ascent 2025 highlighted new business models and workflows when AI agents become more prevalent. The firm sees the biggest value shifting to the application layer, especially with outcome-based pricing replacing traditional software licenses. Leaders must adapt to a world where AI decisions are probabilistic, not certain, and move quickly to capture emerging opportunities. Although the viewpoints might not be hugely differentiated, the presentation videos are worth checking out.
AI-First Businesses: When AI Becomes the Default
In recent weeks, I've been watching with fascination as more companies declare themselves AI-first rather than merely AI-enabled. This shift is more profound than it might initially appear.
In a bold memo titled Reflexive AI usage is now a baseline expectation at Shopify, Shopify CEO Tobi Lütke announced a policy requiring teams to justify the need for human resources by first demonstrating that a task cannot be handled by AI. Employees must incorporate AI into their workflows and are evaluated on their proficiency with AI tools in performance reviews. This directive is enforced uniformly across all levels of the company, including executives.
Frankly, I don't think it's feasible to opt out of learning the skill of applying AI in your craft; you are welcome to try, but I want to be honest I cannot see this working out today, and definitely not tomorrow.
— Tobi Lütke, Shopify CEO
Days later, Duolingo's Luis von Ahn followed with his own announcement that being AI-first means "we will need to rethink much of how we work. Making minor tweaks to systems designed for humans won't get us there."
How Are AI-First Businesses Different from AI-Enabled Ones?
Being AI-enabled means adding AI tools to existing processes—like giving workers a more powerful calculator. The business model and workflows remain fundamentally unchanged; AI is a helpful assistant, not the architect.
Being AI-first means reimagining the entire organization with AI as the foundation. It's about asking: "If we started today, with AI as our foundation, how would we design this business?"
In the CFO suite, for instance, we are starting to hear people question whether the accounting and planning cycles need to exist at all when data and decision silos are broken down by AI. This is AI-first thinking in action.
Shopify's vision also flips the traditional resource allocation model: the default is no longer "hire people and give them tools" but rather "deploy AI and supplement with people where necessary."
The evidence of AI's impact is clear and that’s the motivation behind these companies’ strategic shifts. Duolingo launched 148 new AI-created language courses in a year—a task that previously took them 12 years to achieve for their first 100 courses. Meanwhile, Shopify claims properly used AI can produce "100x the work" compared to traditional methods.
Small Companies Could Be AI-First Too
Smaller businesses are not left behind. Rather, they might actually be better positioned to go AI-first than large enterprises. The rapid advancement in AI agents serves as a strong enabler for organizations with fewer legacy constraints.
Cognition AI is a startup that developed Devin, the world’s first AI software engineer capable of independently planning, coding, debugging, and deploying software projects. Cognition has achieved remarkable scale: Devin is now used in production by major enterprises, the company has partnered with Microsoft, and its valuation soared to $4 billion by March 2025. And yet, the team only has 65 people.
Cognition’s internal structure and philosophy exemplify the AI-first business model. As CEO Scott Wu described on Lenny's Podcast recently, every engineer at Cognition is empowered by a “team” of six Devin AIs working alongside them. The company doesn’t just use AI to assist with tasks; it has reimagined its workflows, roles, and output around AI as the core engine of value creation.
And it’s not just for highly technical teams. Many small businesses in traditional industries could be AI-first too.
Take Perceptis, a new consulting firm built by Alibek Dostiyarov, a former McKinsey consultant, and Yersultan Sapar, a former Apple engineer. Rather than building a traditional firm with armies of analysts, they designed their business so AI agents handle much of the core consulting work: drafting proposals, conducting research, and automating repetitive tasks. The human team focuses on high-value client engagement and strategic oversight. I’ve seen many entrepreneurs in high-value services businesses adopting similar organizational decisions too.
The barriers to entry—data infrastructure, access to AI platforms, and technical know-how—are lower than ever, thanks to cloud-based tools and accessible AI services. What's required isn't so much technical muscle as a willingness to rethink, experiment, and lead with vision.
The Human Element
A persistent fear is that AI-first means fewer jobs and less human involvement. The reality is more nuanced. In AI-first organizations, human roles shift toward what people do best: creativity, critical thinking, relationship-building, and strategic oversight. AI takes on the repetitive, scalable, and data-intensive tasks.
As von Ahn put it: "This isn't about replacing Duos with AI. It's about removing bottlenecks so we can do more with the outstanding Duos we already have. We want you to focus on creative work and real problems, not repetitive tasks."
From a board or executive standpoint, this shift requires careful consideration: how does your company maintain its values and workforce trust while accelerating productivity through AI? And how are strategic talent decisions like hiring, development, and restructuring being governed in light of AI's growing role?
What It Takes to Be AI-First
Before rushing to declare your organization "AI-first," consider what made this approach viable for companies like Shopify, Duolingo, Cognition, and Perceptis:
Data readiness is non-negotiable. Organizations must conduct thorough data audits before implementing AI strategies to identify silos and integration opportunities.
Governance and culture matter as much as technology. Without strong governance, clear strategy, and a culture that embraces experimentation, even advanced AI tools will underperform. For AI-first companies, directors and leaders must ensure that oversight frameworks keep pace with how deeply AI is embedded into operations.
Comfort with imperfection is needed. Lütke's memo specifically states they "would rather move with urgency and take occasional small hits on quality than move slowly and miss the moment."
Leadership alignment across all levels is critical. This isn't a side project but a fundamental strategic priority that requires champions throughout the organization.
Transparency about the journey is paramount. Successful leaders are candid about implications for workers and realistic about challenges. Lütke's public memo is noteworthy precisely because of its clarity.
Investment in skills development must be concrete. Both Shopify and Duolingo emphasized their commitment to upskilling.
Focus on value creation gives direction. The most successful AI-first organizations clearly articulate how AI creates business value.
Technology alone doesn't make a company AI-first. Success requires robust data practices, ethical frameworks, and a culture that fosters agility, encourages rapid iteration, and ensures AI is aligned with business goals and values.
AI Writing Strategies: Improving Your Process and Influencing the Models
Many of us now use AI as a research and brainstorming companion and an editor in writing every day. If you have always been looking for ways to further improve our process, here are some ideas. I’d love to hear yours!
How I break through revision loops that go sideways
After using AI for my basic writing needs, sometimes I found myself in the revision loop. Multiple revisions created different drafts, not better ones. I'd write something, ask AI for improvements, and repeat, only to end up with several versions that each had good parts but none felt quite right. It was like shuffling cards rather than building something cohesive.
What finally worked was changing how I gave instructions. I created a simple quality assurance checklist for the AI to hold up its drafts against, limited edits to specific sections while preserving others, and shared examples of writing I liked. Instead of asking for vague improvements, I explained exactly why previous drafts weren't clicking. I also asked AI to flag parts that might sound off to readers, catching tone issues I often miss in my own writing. When I am happy with the revisions, I will ask AI to summarize key findings through our conversations into writing instructions for future projects.
How Farhad Manu uses AI for nuance and structural feedback
Farhad Manu is a former columnist for the New York Times, and he is described as "one of the most interesting voices in tech writing out there".
In this interview on “How I AI with Claire Vo” Podcast (below), he shares that is treating AI as the "best thesaurus". It goes beyond simple synonyms, helping him find alternative words, discuss their "shades of meaning," and even explore and find idioms and metaphors. He notes it can categorize word suggestions by intent (e.g., more dramatic, colloquial, ironic).
Another interesting approach is how Manu uses AI as an integrated first reader for structural feedback. He gives AI sections of his draft (e.g., the first six or seven paragraphs) to get early feedback on structure. He asks questions like whether his point comes across quickly enough or if there's unnecessary commentary. While he notes it's not sophisticated enough to find logical inconsistencies, it helps polish the writing and integrate feedback during the writing process, not just at the editing stage.
How Tyler Cowen uses AI as secondary literature and intentionally write for AI
Economist and writer Tyler Cowen shared his approach to working with AI on “How I Write” podcast (below). He uses it not just for writing but as his "new secondary literature" to enhance his thinking before podcasts or to deepen his reading. He finds he gets better results with specific questions rather than broad ones. Another tip from Cowen is to regularly try new AI models even though you might think you like a particular one, believing there's no substitute for direct experience as these tools evolve.
“Writing for AI” is the most profound insight from this conversation for me.
Cowen explains that by producing a large and consistent body of work-hundreds of podcasts, daily blog posts, and numerous books-he is effectively creating a rich dataset for AI to learn from, ensuring that these models develop a detailed and accurate representation of his ideas and style. He notes that his writing changes “a bit” with this in mind: he aims for clarity and authenticity, knowing that AI will reference and model his work, sometimes even better than most human readers.
Happy Mother’s Day to all the mothers and mother figures in our lives! We love you.
Thanks for reading,
Joyce