At the Berkeley Haas AI Conference, I heard dozens of takes on AI adoption. Most were either breathlessly optimistic or cautiously vague. The most useful framework came from a VP at Snowflake, who shared the three-pillar model they use internally to decide when and how to apply AI across a 9,000+ person organization.
The Framework
It's deceptively simple. For any task or process, ask which category it falls into:
1. Repetitive? Automate it.
If you're doing something repetitive and rote, find a way to automate it entirely. AI handles the work. You move on.
2. Creative? Augment it.
If the work is creative — writing, design, strategy, analysis — AI does some of the work and you do some of the work. But you're ultimately creating the end product. Human + AI > either alone.
3. Judgment and social intelligence? Stay human.
If it involves emotional intelligence, social connection, or nuanced judgment, don't use AI. Period. "That's the way we're going to stay human, all the way through this process."
The example for "stay human" was telling: when you want to give someone a promotion, talk about their performance, or engage with them personally — don't use AI to craft that communication. It comes from you as a human connecting with another human.
Why This Framework Works
Most AI adoption frameworks I've seen are either too abstract ("use AI strategically") or too specific ("use Claude for X, GPT for Y"). This one operates at exactly the right level of abstraction — it gives you a decision rule for any situation without prescribing specific tools.
It also avoids the two most common mistakes companies make with AI:
- Trying to automate everything, including things that require human judgment (leading to bad outcomes and employee distrust)
- Not automating enough, applying AI only to "safe" creative tasks while ignoring the massive efficiency gains in repetitive work
Real-World Examples
Automate: Job Descriptions
Snowflake built an internal tool for generating job descriptions. The old process: 1-1.5 hours per posting — writing, reviewing previous JDs, manager review, multiple rounds. The new process: enter the job title and a few inputs, the tool searches internal and external data, and produces a polished JD in under 5 minutes. The speaker claimed they've already saved two years of cumulative people-time since launch.
Augment: Mentorship Matching
They built an internal peer-to-peer mentorship program where AI handles the matching — analyzing profiles, preferences, and goals to pair mentors with mentees. The AI does the heavy lifting of finding good matches at scale, but humans still own the actual mentoring relationships.
Stay Human: Leadership and Communication
The hard line: promotions, performance conversations, personal engagement. Even interview feedback, to some extent — while AI can help structure the process, human judgment in evaluating candidates remains irreplaceable. "Leadership is absolutely a durable skill. An AI is still pretty far from motivating, inspiring, influencing, persuading."
Making AI Adoption Non-Scary
One of the biggest barriers to AI adoption isn't technical — it's psychological. People are afraid AI will replace their jobs. The Snowflake team tackled this head-on with three steps:
Step 1: Education. A 15-minute prompt engineering class for all employees. Not technical. Just "how to talk to whatever natural language tool you have." Everyone takes it, practices, and then they work through exercises in groups.
Step 2: A comprehensive curriculum. They created "AI for Everyone" — 10 internal courses covering AI concepts, Snowflake's specific products, and the history of AI. Live classes plus recorded sessions.
Step 3: Start with one thing. After the training, every employee is asked to find one thing in their workday they can automate. Just one. They report back at the end of the week. This creates a virtuous cycle: once people see a small win, they look for the next one.
"Once they learn academically, once they have the framework, and once they actually play with the tools, it becomes much easier for adoption."
The Durable Skills That Survive AI
Both speakers on this panel agreed on what will endure as AI handles more technical work:
- **Leadership** — AI can't motivate teams or navigate organizational politics
- Judgment — Navigating complex, gray-area decisions that require weighing competing interests
- Curiosity — Not just using AI tools, but compulsively exploring new ones. "Did you on a weekend say, 'I'm gonna download this and play with it'? That natural inclination to be curious, to pull threads, to experiment — that's gonna be super important."
- Performance mindset — Openness to change, willingness to embrace new tools, resilience when things shift
My Takeaway
The companies that will win AI adoption aren't the ones with the most sophisticated technology. They're the ones that give their people a simple framework for deciding what to automate, what to augment, and what to keep human — and then make it psychologically safe to experiment.
Automate the repetitive. Augment the creative. Stay human where it matters. It's the simplest useful framework I've encountered for navigating the AI era.