Learning & Growth in the AI Age
The Self-Taught Journey
Holly Smith’s career path challenges the myth that success in tech requires a perfect linear progression. Starting with a math degree and roles at Lloyd’s Banking Group, she admits to being “very much on the back foot” when joining Databricks. Her candid recollection of asking “what’s a JVM?” when learning about Spark reveals a truth many professionals hide: we all start somewhere, and that somewhere is often uncomfortable.
This honesty and vulnerability matters because the pace of change in technology has placed many people in similar positions. As Holly observes, “of all the people doing Gen AI today, two years ago, I reckon like 95 percent of them weren’t doing this.” The playing field has leveled in unexpected ways.
Systems Over Willpower
The conversation shifts to a crucial insight: sustainable learning isn’t about daily heroics. “It’s really not about trying every day,” Holly explains. “You’re going to have days when you are tired, you’re going to be exhausted.” Instead, she advocates for building systems that make learning inevitable rather than aspirational.
This means deliberately choosing roles where learning is embedded, not optional. It also means finding communities, like Slack channels, LinkedIn groups, conferences, that expose you to what you don’t know you don’t know; or in bolder terms, being proactive to avoid the Dunning-Kruger effect writ large: that dangerous peak of confidence where you “don’t know what you don’t know.” Holly’s approach is to actively seek out those gaps through thoughtful and purposed learning and growth positioning rather than pure determination.
The principle of “1 percent a day” emerges as more sustainable than periodic sprints. Continuous, incremental learning beats intensive bursts that, while fun, inevitably fade.
The Language Barrier
One of Holly’s core frustrations is how both finance and technology industries obscure knowledge behind unnecessary jargon. “These are two groups of people who love to put lots of buzzwords and specific language about things,” she notes. This creates artificial barriers to entry that have nothing to do with actual capability.
Her mission in developer advocacy centers on plain English explanations that don’t assume computer science foundations. She regularly sees technically proficient people from all angles. Security experts, for example, struggle when they encounter data work simply because the language and assumptions differ. “The amount of people that come up to you afterwards and they’re like, ‘Oh, Holly, thank you so much. I actually understood that.’ I sat there thinking, hang on a minute. I thought you were smarter than me.”
This insight extends beyond individual communication to organizational lexicons. Even defining something as seemingly simple as “net margin” can confound companies internally. If businesses can’t agree on basic financial terms, how can technical teams expect shared understanding of terms like “COE” or “SIG” without explicit definition?
Culture as Permission Structure
The discussion reveals how organizational culture manifests in attitudes toward learning and experimentation. Some companies have “tolerance for inertia” or a spectrum from those paralyzed by risk to those chasing every shiny object. Neither extreme serves us well.
Holly identifies specific cultural markers:
Punishing uncertainty: In some environments, saying “I don’t know, let me research that and get back to you Monday” is met with horror rather than professionalism. These are cultures that confuse confidence with competence.
Waterfall rigidity: Organizations that demand complete certainty upfront, with no tolerance for iteration as understanding develops, create hostile environments for learning.
Resource starvation: When everyone operates at 120 percent capacity year-round, learning becomes impossible. When a company won’t expense an eight-pound book for an employee willing to invest their personal time reading it, the message is clear.
Conversely, healthy learning cultures provide:
Multiple learning modalities (formal training, conferences, self-paced courses, evening classes)
Psychological safety to admit knowledge gaps
Time and budget explicitly allocated to development
Leadership that models curiosity rather than omniscience
The Executive Blind Spot
Holly shares observations from senior friends about what separates effective executives from ineffective ones when facing AI transformation: humility. The best executives ask their teams fundamental questions: “Why is this taking so long? What are the impacts if we get it wrong? If there’s a tiny margin of error, can we spend less time on it?”
The worst declare “I’m not a data person” and shut down. As the conversation notes, this denial is almost absurd as humans process astonishing amounts of information per second through “wetware statistical processes.” Refusing to engage with data is refusing to acknowledge how our brains actually function.
This pattern extends to the common mistake of sending a few people to training and expecting them to return as perfect trainers for the entire organization. “Just cause you’re a practitioner does not mean that you’re going to be great at teaching other people,” Holly observes. She herself spent months shadowing courses and doing dry runs before delivering training at Databricks. The assumption that expertise automatically transfers to teaching ability sets everyone up for failure.
Anti-Patterns and Bike Shedding
“Bike shedding”, the concept where groups tasked with building nuclear power plants spend all their time designing bike sheds because they actually understand those, captures a common failure mode. This typically emerges in “low confidence environments” where no one feels equipped to tackle the real challenge, so they busy themselves with comfortable peripherals while convincing themselves of progress.
The ultimate anti-pattern is wholesale outsourcing of data work. While consulting has its place (a point Holly acknowledges from her own consulting background), treating data as something to be entirely handed off reveals fundamental misunderstanding. “Data is everything,” as noted in the conversation. “It’s your history, your company, the history of your interactions.” Why would you cede that to external parties?
Historically, technology leadership meant outsourcing whatever possible as cheaply as possible. That worked when technology was simpler and less strategic.
Tools and Practical Approaches
Modern AI tools have transformed learning possibilities. Holly specifically mentions Perplexity AI for constant questioning and code editors with AI assistance as legitimate accelerators. However, she warns against pure self-direction such as giving someone a LinkedIn Learning subscription with “hundreds of thousands of hours of content” but no curation or guidance typically fails.
The balance lies between excessive structure (only one approved learning path) and complete chaos (all of the internet, good luck). Effective approaches provide guide rails while respecting individual learning styles.
For hands-on learners, she advocates taking an afternoon with documentation open and experimenting. For executives, it’s about asking the right questions rather than learning to code. The key is matching approach to role and learning style while maintaining consistent forward momentum.
The Sideways Move Strategy
Rather than only thinking in terms of promotions or dramatic career changes, Holly advocates for sideways moves wherein you take on different projects or move laterally within organizations. “A sideways move is always so much easier,” she explains. You can speak openly, you already understand organizational context, and you’re more likely to find innovation happening somewhere in a larger company than you might expect.
This especially matters for AI work, where the experience curve compresses dramatically. Someone with 18 months of LLM experience can legitimately be considered highly experienced because the field itself is so new. “You’re not competing with people that have been doing this for like over a decade,” Holly notes.
The Accountability Question
The conversation concludes with a practical framework: look at what the organization holds people accountable for. If metrics are purely financial, learning likely isn’t valued regardless of stated policies. If accountability includes broader measures, learning can thrive.
The key questions become:
Does my desired learning align with organizational priorities?
Does the company reward or punish learning attempts?
Am I in the right environment for my current life stage?
It’s acceptable to be in a low-growth phase if other life priorities dominate. The critical element is honest self-assessment and alignment rather than forcing mismatched expectations.
Final Thoughts
Learning in the AI age isn’t fundamentally different from learning in any era of rapid change, rather it requires systems over heroics, communities over isolation, and psychological safety over performative confidence. What has changed is the compression of experience curves and the universality of novice status in emerging domains.
As Holly demonstrates through her own journey, expertise isn’t about never asking “what’s a JVM?” It’s about building systems that ensure you’ll eventually understand not just JVMs but whatever equivalent emerges next. The specific knowledge becomes less important than the infrastructure for continuous learning.
The organizations and individuals who thrive will be those who recognize this fundamental shift and build accordingly, not through proclamations about being “learning organizations,” but through concrete actions: time allocated, resources provided, uncertainty tolerated, and curiosity rewarded at every level.
This blog post is based on a conversation between Sid Atkinson and Holly Smith on the Data Culture Podcast. Holly Smith is a Staff Developer Advocate at Databricks. Connect with her on LinkedIn by searching “Holly Smith Databricks.”