Most leaders think they’re good communicators. Fernando Garcia Valenzuela decided to find out.
Fernando is the Head of Cloud Storage Engineering at Atlassian, a 70+ person org spanning three time zones. Staying connected across a team of that scale is no easy feat.
“Warmth doesn’t scale automatically,” he explains. “Empathy doesn’t survive busyness unless you’re deliberate about it. I wasn’t willing to let ‘I have a lot on my plate’ become an excuse for people not feeling seen.”
Fernando needed a way to communicate across his entire team without abandoning his standards. When he couldn’t find one, he built it.
research at a glance
The experiment: A senior engineering leader at Atlassian built a personal AI “pit wall” to audit his own Slack communication across a 70+ person org spanning three time zones.
The method: He created AI agents that analyze his own sent messages, flag missed moments of recognition, and suggest rewrites.
The top 3 lessons learned:
- You can’t fix what you can’t see — and most leaders can’t see their patterns.
- The hardest part about creating a useful agent is defining what “good” looks like.
- Your leadership style isn’t finished. Keep improving it like you would any product.
The setup: scaling fast without losing connection
Fernando is a skilled context-switcher and a fast, present communicator — skills he’s spent years building. When his responsibilities and team expanded, he wanted to keep that speed without losing warmth.
Async writing strips out tone, body language, and the social cues that make people feel seen. The faster you write, the more of that connective tissue falls away. Fernando knew this intellectually, but wasn’t sure how it was playing out in his own day-to-day messages.
The build: a “pit wall” of agents
When building his AI feedback system, Fernando drew on a passion for Formula 1 racing. The pit wall — where strategists relay real-time data to drivers mid-race — became his mental model for how an AI agent should surface insights: timely, contextual, and actionable without pulling the driver off the track.
The agent names followed suit:
Smedley (named for the race engineer famous for “Felipe baby, stay cool”) runs daily. It scans Fernando’s own DMs with direct reports, flags missed opportunities for recognition, and surfaces suggested rewrites — not to replace his voice, but to show him what a warmer version of the same message might look like.
Brundle (named for the F1 commentator known for his sharp analytical breakdowns) runs fortnightly. It pulls across a broader canvas of DMs, peer threads, public channels, and skip-level conversations. Brundle looks for patterns: tone, warmth, and how his communication style shifts by audience. Does he write differently to individual contributors than to peers? Does his warmth drop off on Fridays? Does reassurance land without anything concrete to anchor it?
Fernando built his agents on tools already in his environment — Slack as the data source, Rovo for the AI layer, and Confluence for storing reports — but the pattern is portable to other stacks.
The prompt: The heart of the system
The agents are only as good as the instructions behind them. Fernando’s prompt isn’t a simple command. Instead, it’s a signal taxonomy: a structured definition of what to look for, how to classify it, and what a better version might look like. With a well-built prompt, AI responds with pragmatic advice drawing from world-class communication experts like Teresa Amabile (The Progress Principle), Amy Edmondson (psychological safety), and Kim Scott (radical candor).
An effective prompt includes three things:
- Patterns to improve: Specificity is your ally. Instead of instructing his agents to “be warmer,” he instructed it to flag things like single-word confirmations to individual contributors and missed recognition after a win.
- Context that changes interpretation: Context that seems obvious to you should be detailed in the prompt. Consider how communication differs between public channels versus private DMs, high-stress moments versus routine check-ins, and in skip-level meetings versus communication with peers.
- Output format: Fernando instructed his agents on exactly what format he preferred. Instead of lengthy prose, he receives pattern summaries, signal breakdowns broken into priorities, and links to Slack messages that may require follow up.
Lessons for enterprise leaders
Fernando’s experiment is still running. But the results so far have already reshaped how he thinks about AI, leadership, and what improving actually looks like in practice.
Small signals compound faster than big failures.
“I expected to catch occasional tone misses in high-stakes conversations,” he says. “What I didn’t expect was how consistent the small things were: single-word confirmations to individual contributors, subtle differences in how warmth and recognition showed up across audiences and situations, and reassurances I was making without anything concrete attached.”
None of these are dramatic failures. But across hundreds of messages and dozens of relationships, they shape whether people feel seen or overlooked. The value was less about catching a crisis and more about surfacing the drift he’d never notice on his own.
Defining what good looks like is the real work.
Early versions of the pit wall flagged messages that in isolation seemed off, but in context were perfectly fine. For example, one peer exchange that the AI originally flagged as venting turned out to be appropriate professional escalation.
The fix was more clarity to tell the AI what to look for. Fernando had to get specific: What counts as recognition? Does tone matter differently by seniority? Should it flag a single message or only a pattern? Each question forced him to articulate standards he’d never written down for what good communication actually looks like across his team.
Knowing you’ll review changes how you write.
Beyond the fortnightly reports, Fernando noticed something he hadn’t anticipated: the act of knowing his agents would audit his messages changed how he composed them in real time. He started pausing before sending — not because AI told him to, but because he’d internalized the patterns it surfaced.
He also added a new module to his pit crew that evaluates messages before he sends them out. It provides on-demand editorial help available at any hour, across any timezone, with full context on what he’s trying to improve.
Real examples from Fernando’s AI “pit wall”
Fernando’s reports flag specific messages with specific suggestions. Here are two real examples, anonymized.
🟡 A coaching moment (Smedley): A direct report sent a well-organized message confirming project dates. Fernando replied with clear direction — but focused entirely on the correction and missed acknowledging the planning work they’d put in. Smedley’s suggested rewrite: “Thanks for getting this organized — sounds good. Let’s say…” These five extra words conveyed the same information with a very different signal.
🟢 A positive (Brundle): A direct report replied to an incident query on a public holiday in their timezone. Fernando’s response acknowledged the holiday, apologized for the imposition, and tied the appreciation to their specific effort — not a generic “thanks.” Brundle flagged it as a model worth repeating.
The signals most leaders miss
As Fernando’s experiment shows, the highest-value use of AI may not be doing more work faster. It may be helping leaders see the parts of their work they normally miss: the recognition they skipped, the tone that shifted under pressure, the small signals that shape whether a distributed team feels connected or alone.
Fernando’s already building the next layer — agents that factor in org changes, tenure, and real-time nudges before he hits send. The system isn’t finished because the question it answers never is.
The State of Teams research from Atlassian’s Teamwork Lab found that 89% of executives say AI is making work faster — but only 6% can point to clear, org-wide ROI. That gap exists because most AI deployments optimize for speed, not for how leaders and teams actually work together. Fernando’s experiment points to where the real ROI might live: not in replacing judgment, empathy, or trust — but in giving leaders a mirror they can actually use.
As Fernando puts it, “You don’t need a perfect system. You just need to start looking.”
