Navigating AI + Team Effectiveness with Llewellyn Falco: Highlights from SEPTalks

At a recent SEP Talks event—“AI, Team Effectiveness & the Future of Software Development”—we hosted a live fireside chat with industry expert Llewellyn Falco. We invited Llewellyn to help us think more deeply about the impact of AI on how we build software, collaborate, and grow as teams. What followed was a series of sharp insights and mindset shifts that challenged how we approach hiring, learning, and adapting to change.

Here are a few takeaways that stuck with us—ideas we’re still talking about as we rethink how we work in this AI-powered world.

1️⃣ The Unprecedented Pace of Change

Software development has never moved this fast. Llewellyn summed it up best:

It’s just gotten faster and faster—exhausting, really. I feel exhausted just keeping up with everything that’s coming out.

Most of us know things aren’t slowing down, but hearing it framed this way underscores why we need new approaches: you can’t simply “keep pace” by working harder. Recognizing that relentless acceleration is the new normal helps explain why we need smarter workflows, constant learning loops, and AI as a partner—not just another tool.

Keeping this in mind is critical because many of the practices, tools, and team structures we’ve relied on were built for a more predictable pace. When the ground is shifting this quickly, yesterday’s best practices can become today’s bottlenecks. To stay relevant and resilient, teams need to intentionally evolve—not just react. That means investing in adaptability, prioritizing experimentation, and creating space for people to learn and improve continuously. It’s not about bracing for disruption—it’s about building for it.

---

2️⃣ Where Teams Stand on AI

In our conversation, Llewellyn Falco talked about the different experiences people may have using AI.

He broke teams he frequently encounters into four groups:

• Building AI products from the ground up, where AI is the core.
• Embedding AI features into existing apps.
• Using AI to accelerate development, even if the shipped product isn’t “AI-driven.”
• Heard of AI—most people, who haven’t yet explored deeper possibilities.

Understanding which category your team falls into is crucial because each path demands a different mindset, strategy, and set of tools. Building AI from the ground up requires deep technical expertise and a focus on innovation, while embedding AI into existing products often means balancing enhancement with user experience. Using AI to accelerate development is more about workflow optimization and productivity gains than product transformation. And for those just beginning to explore AI, the priority is education and experimentation. Knowing where you stand helps set realistic goals, allocate resources wisely, and avoid mismatched expectations across teams and stakeholders.

---

3️⃣ Turbocharging Learning Through Collaboration

So how do you keep learning when the ground keeps shifting? According to Llewellyn Falco, the answer is simple but often overlooked: collaboration. Practices like pair programming and mob programming aren’t just about sharing knowledge—they disrupt routine thinking. They force you to engage, adapt, and learn in real time.

Solo study has its place, but it’s easy to fall into familiar patterns—repeating what you already know, even when it’s no longer effective. It’s like always ordering the same dish at your favorite restaurant: comfortable, but not growth-inducing. Collaborative practices push you out of that comfort zone.

Why does this matter? Because learning isn’t just about consuming new information—it’s about changing how you think. Pairing and mobbing make that change visible and immediate. They expose blind spots, introduce new techniques organically, and accelerate skill-building in ways self-paced learning often can’t.

If your team adopts even a slice of this mindset—rotating review partners, pairing across domains, participating in community coding sprints—you’ll find that collaboration becomes your most powerful engine for learning and adaptation.

---

4️⃣ Keeping Code Maintainable in an AI Era

Code quality and maintainability aren’t just best practices—they’re force multipliers when working with AI tools. Clean, well-structured code allows AI systems to better understand, refactor, and extend your codebase with minimal human intervention.

Llewellyn Falco has seen AI migrate or modernize massive legacy systems in hours—something that once took weeks. But the success of these transformations often hinges on how readable and consistent the original code is.

When code is tangled or poorly documented, even the most advanced AI struggles to make sense of it. By investing in maintainable code today, teams set themselves up to fully leverage AI tomorrow—whether for modernization, automation, or scaling development efforts.

---

5️⃣ Cheap, Fast Feedback Loops

Using AI in software development can feel risky—especially when code quality is critical.

AI tools are not deterministic by nature, which means they may produce different results even with the same input. We often don’t fully understand how or why they make certain changes. That unpredictability can be unsettling, particularly when you’re building safety-critical systems like medical devices where lives are at stake.

That’s why it’s essential to rely on fast, low-cost feedback loops and run small experiments.

These help teams safely explore how AI behaves, build trust in its outputs, and better understand where it adds value—before committing to widespread use.

To experiment safely, you need feedback loops that are both cheap and fast:

• Make small, reversible changes.
• Commit early and often (e.g. what tests are passing).
• Revert what fails without hesitation.
• Automate canary releases or rollbacks so mistakes don’t take down your whole system.

Llewellyn emphasizes that the value of a feedback loop isn’t just in speed—it’s in minimizing the “blast radius” of mistakes so you can iterate relentlessly:

I’d take one file at a time, ask the AI to convert it to TypeScript, paste it back, run tests, and—if it passed—commit; if it failed, I’d revert immediately. Mistakes were cheap, and every successful commit felt like a win

Why “cheap” matters:

Lower risk of failure – A small mistake in a 50-line file can be discarded in seconds, whereas a tangled, multi-module change can consume hours or days to untangle.

Encourages experimentation – When reverts cost almost nothing, you’re free to explore bold ideas rather than defaulting to safe, familiar approaches.

Reduces cognitive load – Tiny changes keep your mental model tightly aligned: you only need to remember one diff rather than an avalanche of interdependent edits.

Why “fast” matters:

Shortens learning cycles – The quicker you see a test result or user reaction, the faster you learn what works (and what doesn’t).

Maintains momentum – Waiting 30 minutes for a full-suite run or manual review breaks your flow; sub-minute feedback keeps your focus razor-sharp.

Boosts confidence – Frequent green builds reinforce progress, while instant reverts avoid the dread of lengthy rollbacks.

In practice, even with AI speeding up transformations, you still need to:

• Break changes into tiny, reversible steps
• Automate tests and deployments (canary releases, rollbacks)
• Commit successes immediately and discard failures

By making each iteration both low-cost and rapid, you transform every trial into a meaningful learning opportunity—and keep the forward momentum alive.

---

6️⃣ Cultivating a Growth Mindset

Llewellyn Falco cautions that relying too heavily on AI can lead to a “fixed mindset”—where we assume that AI is always right and stop engaging deeply with the work. But to truly innovate, humans must stay at the center of the process. AI is a powerful tool, not a replacement for critical thinking. When we blindly accept its outputs, we risk:

• Overconfidence: AI can produce convincing but flawed solutions. Without human scrutiny, bugs slip through and learning stalls.
• Shallow understanding: Letting AI write every line bypasses the struggle that builds real expertise. Mastery comes from curiosity, not shortcuts.
• Fear of failure: If success feels automated, mistakes can feel personal. But when we see errors as chances to grow—both in prompting and problem-solving—we build resilience.

Drawing on Carol Dweck’s research, Falco reminds us that praising effort and curiosity—not just intelligence—fosters a growth mindset. In an AI-powered world, the most effective teams will be the ones that stay curious, keep asking “why,” and use AI to amplify—not replace—their own thinking.

---

7️⃣ A Better Way to Hire

Llewellyn shared how AI is reshaping the traditional hiring process—and how the old methods are falling short. After years of sorting through resumes, he realized, “I couldn’t figure out a single useful thing from a resume ever,” and eventually stopped reading them altogether. Instead, he created a lightweight coding test to decide whether to move forward with a candidate—and consistently uncovered talented people who would have otherwise been overlooked.

It was a good reminder of why we (SEP) take a different approach to interviewing. Résumés tell you about the past. We’re more interested in how someone thinks now and how they’ll adapt in the future.

In this new AI-powered era, it’s tempting to zero in on whether someone knows how to prompt ChatGPT or use the latest tool. While AI literacy does matter, it’s not the whole picture—and over-focusing on tool familiarity can cause you to miss the bigger signal. The pace of change means today’s AI tool may be obsolete tomorrow. What endures is a candidate’s ability to think critically, adapt quickly, and solve problems creatively.

At the same time, AI has changed how we solve problems. It’s not enough to assess whether someone can debug a loop or build a feature from scratch. We now need to know: can they leverage AI to extend their capabilities? Can they collaborate with it, question it, and refine its output? Or are they still trying to solve 2025 problems with a 2015 mindset?

Hiring in this new world means looking for developers who are not just technically capable, but mentally agile—people who are curious, collaborative, and ready to work with AI as a thinking partner. That’s the skillset that will stand the test of time.

---

8️⃣Embrace Cheap Experiments

Finally, don’t be afraid to try lots of small experiments—and toss the failures away. It’s like exploring new walking routes: most lead nowhere special, but every once in a while, you discover something amazing. As long as each experiment is low-cost and easy to roll back, those breakthroughs will keep you at the forefront.

Llewellyn stresses that breakthroughs often come from running many small, low-cost experiments—then tossing out everything that doesn’t work. Two vivid examples from our chat:

Exploring Every Street

During the pandemic I started daily walks with a friend—our only rule was never to take the same route twice. In six weeks I’d walked every public street in Los Alamos. Most days nothing special happened—but the few times I stumbled on an old neighbor’s story, a hidden garden, or a quirky house paint job, those discoveries were gold.

Because each walk cost the same effort, Llewellyn could afford dozens of “blank” days. The rare, delightful finds more than justified the handful of wasted steps—and his mental model of the town expanded dramatically.

Code Transformations by AI

I needed to migrate a large JavaScript project to TypeScript. With AI, we got 95 % converted in three hours—something that would’ve taken me two to three weeks by hand. I’d feed one file at a time to the model, paste back the result, run tests, commit if it passed, or immediately revert if it failed. Mistakes were cheap, and every successful commit felt like a win.

That same cycle—transform → test → commit/revert—can be applied to refactoring, dependency upgrades, or any batch code change. By keeping each step small and reversible, you minimize risk while maximizing learning speed.

---

9️⃣ Using AI To Get Your Other Work Done

Llewellyn even applies AI to his daily post-work routine:

1. He prompts ChatGPT with his past daily statuses and asks it to “interview” him for each team’s progress.
2. He answers focused questions instead of staring at a blank page.
3. He iterates on the draft—shortening where needed, tweaking tone.
4. He then asks the AI to suggest an illustrative image prompt, picks a style, and hands it off to generate the graphic.
5. Finally, he asks the AI to critique its own prompt for tomorrow’s run.

By turning routine tasks into guided, collaborative workflows, Llewellyn offloads cognitive overhead and avoids decision fatigue. The result isn’t just saved time—it’s sustained focus, higher-quality output, and more mental energy left for the creative, human parts of the job that AI can’t replace.

Here are a few examples we’ve started using recently:

• Creating status reports
  AI can summarize updates, prompt for missing details, and format the report clearly—turning scattered notes into polished updates with minimal effort.
• Drafting meeting agendas or recaps
  Feed in calendar events or meeting transcripts, and AI can generate structured agendas or concise summaries, freeing up time for actual decision-making.
• Preparing presentations
  AI can help outline slides, suggest visuals, and even generate speaker notes based on raw content or goals.

---

Llewellyn helped remind us that true progress comes from collaboration, rapid experimentation, and treating AI as a partner in learning. Here’s to harnessing those ideas in your own work!

---