One morning, I watched a data engineer struggle with using AI for thirty minutes, trying to debug a DBT job. The problem wasn’t the LLM’s capabilities—it was how the engineer framed the question. No context about what they’d already tried. No explanation of the expected versus actual output. Just “fix this code” followed by a massive code dump.

This same engineer had similar struggles with stakeholders. Presentations that assumed too much context. Emails that buried the ask. Meetings where they answered questions nobody asked.

Here’s what I realized: the framework for good human conversation is the same framework for effective AI interaction. Whether you’re prompting an LLM or presenting to executives, the mechanics are identical—you’re coordinating with another entity to exchange information and achieve shared goals.

Harvard professor Alison Wood Brooks spent years analyzing thousands of conversations, discovering that communication follows predictable, learnable patterns. Her research culminated in the TALK framework—a systematic approach to conversation that works whether you’re talking to Claude, Chat GPT or your CEO, or a date you’re not sure about. For technical professionals who recognize communication matters but don’t feel it’s their natural strength, this is a game-changer.

The research is clear: 70% of career advancement for engineers stems from communication skills, not technical prowess. Yet we spend hours learning data optimization and zero minutes learning conversation science. Brooks’ findings offer evidence-based techniques that play to analytical strengths—asking better questions beats being charismatic, preparation trumps improvisation, and introverts have hidden advantages.

This isn’t about becoming an extrovert. It’s about understanding how conversation actually works and deploying specific, research-backed strategies. The same strategies that make you better at prompting AI make you better at managing stakeholders, building teams, and advancing your career.



T Is for Topics: Preparation Isn’t Performance


Most people balk at preparing conversation topics. It feels unnatural, performative, like you’re planning spontaneity out of existence. Brooks found that 53% of people think topic prep is unnecessary, while 50% believe it will decrease their enjoyment of conversations.

This resistance stems from what Brooks calls the “Myth of Naturalness”—the belief that charismatic people just know what to say. We see someone excel at conversation and assume it’s effortless for them, when in reality they’ve accumulated years of deliberate practice. Same pattern we see in technical skills, but we forget to apply the logic.

Here’s the data: before conversations, 27% of people spend at least five minutes deciding what to wear, while only 18% think about what they’ll talk about. We optimize our appearance and neglect the substance. Then we wonder why conversations feel awkward or shallow.

The System 1 vs System 2 Problem

During conversation, we rely on “system 1 thinking”—fast, instinctive, automatic processing. This guides us toward whatever’s top of mind: the weather, that random person standing nearby, the fact that someone just ordered appetizers. We talk about what’s accessible rather than what’s meaningful.

The topics most accessible to us aren’t always best for connection. We drift into safe small talk rather than searching for mutually rewarding subjects. We prove how smart we are instead of learning what our partner knows. This is natural egocentrism—optimizing for ourselves rather than the coordination game.

Topic preparation engages “system 2 thinking”—slower, deliberative, logical processing. Before the conversation, when you have mental space, brainstorm 3-5 topics aligned with your goals. This isn’t scripting—it’s cognitive offloading. By reducing information-processing requirements beforehand, you free mental space during the conversation to actually listen and respond.

What This Looks Like for Data Engineers

Before that stakeholder meeting, don’t just prepare your technical update. Prepare topics:

  • “I’m curious what drove the timeline pressure you mentioned last week”
  • “What would success look like from your team’s perspective?”
  • “How does this fit with the Q2 roadmap priorities?”

Before the networking event, prepare genuine curiosities:

  • “What made you move from engineering into product management?”
  • “What’s the hardest part of scaling data infrastructure at your company?”
  • “How did your team approach the migration to Snowflake?”

For your 1:1 with your manager, beyond status updates:

  • “Where do you see the biggest risks in our current architecture?”
  • “What capabilities do you wish the data team had?”
  • “How can I better support the team’s strategic goals?”

Brooks uses a “topic pyramid” framework—base level is weather and small talk, middle level involves experiences and perspectives, top level consists of topics only you and your partner could discuss in that moment. The mistake most people make: getting stuck at the base. Small talk isn’t the problem; lingering there too long is.

Prepare topics that climb the pyramid. Instead of “How was your weekend?” (base level), try “I’ve been curious what made you interested in data quality originally” (middle to top level). You’re not abandoning spontaneity—you’re creating scaffolding that lets meaningful conversation emerge.

How This Applies to LLM Communication

The exact same principle applies when prompting AI. Most people use system 1 thinking—whatever question first comes to mind. “Debug this code.” “Summarize this document.” “Write me a function.”

Better results come from system 2 prep. Before opening Claude, ask yourself:

  • What’s my actual goal here?
  • What context does Claude need?
  • What have I already tried?
  • What specific output format do I want?
  • What constraints matter?

Compare these prompts:

System 1: “Fix this SQL query”

System 2: “I’m getting duplicate rows in this SQL query that joins orders to customers. I expected one row per order, but I’m getting multiple. Here’s the query, sample input data (remove sensitive data/mask it), and current output. Can you identify why duplicates are appearing and suggest a fix?”

The second provides context (what’s wrong), expected behavior (one row per order), actual behavior (multiple rows), and supporting information (query, data, output). This is topic preparation for AI—giving the LLM the topics it needs to help you effectively.



A Is for Asking: Follow-Up Questions Are Your Superpower


Brooks’ team analyzed 900+ speed dates and conversations among 398 strangers, classifying every question asked. They identified four types: introductory (“What’s your name?”), mirror (“I’m good. How are you?”), topic-switching, and follow-up questions.

The finding that should change how you approach every conversation: follow-up questions drive almost all positive effects on likeability.

The numbers are striking. People in the top third of question-askers got the most second dates. Asking just one more follow-up question per date across 20 dates resulted in one additional second date agreement. Mirror questions—the polite reciprocations we default to—produced no measurable increase in likeability.

Why Follow-Ups Work

Follow-up questions signal three things simultaneously: you were actually listening, you care about what the person said, and you’re thoughtful enough to connect their previous statement to a deeper inquiry. They make people feel heard and validated.

When a product manager explains a feature request, the instinct is to immediately propose technical solutions. That’s a shift response—redirecting attention to your expertise. A follow-up question maintains focus: “What’s driving the timeline pressure you mentioned?” This signals engagement far more effectively than demonstrating knowledge.

Brooks warns against “ZQs”—people who leave conversations having asked zero questions. Professional matchmaker Rachel Greenwald says it directly: “Zero questions means zero second dates.” But even when we ask questions, we vastly overestimate how many.

In one study, negotiators estimated that over 50% of their turns included a question. In reality, less than 10% did—less than one-fifth of what they thought. We’ve found the same pattern in friend conversations and first dates. In the chaos of conversation, we don’t notice the link between asking questions and being liked, so we don’t get clear feedback on what’s working.

The Four Question Types (And Which Matter Most)

Introductory questions (“What’s your name?” “How are you?”) are necessary social ritual. Important but don’t linger.

Mirror questions (“I’m good. How are you?”) reciprocate what was just asked. They’re polite but add little value—they’re “low-hanging improvisational fruit” that feel effortless but don’t increase likeability. Use them when you genuinely want to hear the answer, not out of reflex.

Topic-switching questions initiate noticeably new subjects. Instead of mirroring “How was your day?” with your own day summary, try: “Did you see that new data governance framework GitHub just released?” This climbs the topic pyramid rather than floundering at the base.

Follow-up questions are the superheroes. They keep you on the present subject, probing deeper on something your partner said. “You mentioned the migration was stressful—what made it complicated?” “How did your team approach that problem?” “What would you do differently next time?”

In Brooks’ studies, when people were prompted to ask more questions, the extra questions were mainly follow-ups. This means the rewards of asking more questions were driven almost entirely by follow-up questions. They unlock deeper learning, feel personal and validating, and show you’ve actually been heard.

The Never-Ending Follow-Up Exercise

Brooks challenges her students to have conversations using only follow-up questions. They can make statements for smoothness, but must always respond with a follow-up. The goal: see how easy and powerful sustained curiosity is.

Students are surprised by three things:

  1. How easily they can maintain never-ending follow-ups
  2. How much they learn about their partner
  3. How engaging it is—for everyone involved

Great conversationalists strike the right balance between topic-switching questions (to hop to new islands) and follow-up questions (to explore islands once you’re there). Typically, asking mostly follow-up questions doesn’t mean staying on the same topic—they lead your partner through exciting drift, opening new worlds even as they spring from what was previously said.

For Data Engineers: Learning from Masters

Terry Gross, Oprah Winfrey, and Barbara Walters are follow-up masters. They don’t need specialized expertise to discuss any topic because they can ask follow-up questions. It’s the ultimate improv tool, especially when you don’t understand or know what to say.

Specific patterns to emulate:

Emotional probing : “How did that make you feel?” When a stakeholder says “the migration was challenging,” instead of proposing solutions: “What was most frustrating about the process?”

Past-present-future : “If you had to do it again, would you choose Snowflake?” “Are you happy with the decision now?” “What will you tell the next team considering this migration?” This breaks people out of present-myopia and reveals new insights.

Journalistic curiosity : When in doubt, just keep asking. The key to good conversation isn’t knowing—it’s learning. Be interested in your partner, not interesting yourself.

When a colleague mentions they’re struggling with data quality, resist immediately showcasing your expertise. Instead:

  • “What specific quality issues are you seeing?”
  • “How is this affecting downstream teams?”
  • “What have you tried so far?”
  • “What would ideal state look like?”

You’ll learn more, build better rapport, and position yourself as someone who actually listens—a rarity in technical organizations.

The Boomerasking Trap

Brooks identifies a destructive pattern called “boomerasking”—asking questions purely to talk about yourself. “Have you won any contests lately?” followed immediately by “I won the Pick a Present contest!” The question goes out and returns like a boomerang.

In technical contexts, this manifests as:

  • “How did your team handle the migration?” → Immediate pivot to “Here’s how we did ours”
  • “What challenges are you facing?” → Instant “Let me tell you about our challenges”

Sociologist Charles Derber distinguishes between shift responses (redirecting to yourself) and support responses (maintaining focus on your partner):

Shift: “Yeah, I had the same issue on my last project—let me tell you what happened” Support: “What’s making the timeline feel tight?”

For technical professionals, the instinct to demonstrate expertise through personal examples creates precisely the wrong impression. Three types of support responses build trust:

  • Background acknowledgments: “Seriously?” “Oh wow!”
  • Supportive assertions: “That sounds really difficult”
  • Supportive questions: “What happened next?” “How did that affect the team?”

How This Applies to LLM Communication

The best LLM interactions are conversations, not commands. Follow-up with Claude based on responses:

Initial: “Can you help me optimize this SQL query?”

Claude responds with a solution

Follow-up: “Why did you choose a CTE instead of a subquery here? What are the performance implications?”

Claude explains

Follow-up: “In what scenarios would the subquery approach be better?”

This creates a learning conversation rather than a one-shot transaction. You’re not just getting code—you’re understanding the reasoning, which makes you better at writing queries independently.

Same pattern works in code review. Instead of “Why did you do it this way?” (which sounds accusatory), try:

  • “Can you walk me through your thinking here?”
  • “What were you optimizing for?”
  • “Did you consider [alternative]? What made you choose this approach?”

Follow-up questions signal curiosity rather than criticism, making people more willing to explain and learn.



L Is for Levity: Energy Management, Not Entertainment


Brooks introduces levity using an emotion chart with two dimensions: arousal (high vs low energy) and valence (pleasure vs displeasure). When conversations drift into the lower-left quadrant (sad, bored, disengaged), people disengage. Minds wander. Eye contact drops. Words become neutral or negative. Pauses lengthen.

In the upper-right quadrant (excited, engaged, joyous), the opposite happens. People lean forward, make eye contact, give back-channel feedback (“yeah,” “uh-huh”), respond quickly, even interrupt—because they’re excited to know what happens next.

Levity is any conversational move—playful, funny, unexpected, warm—that infuses positive energy. It’s the yeast that turns dense dough into fluffy bread, the helium keeping a balloon aloft. It moves conversations from disengagement to engagement.

Why Engineers Underuse Levity

Research shows people overestimate how often humor goes poorly and underestimate the upside. At age 23, 84% of people report smiling and laughing boosts mood, draws people closer, and increases perceptions of competence.

Brooks’ research found that managers randomly assigned to make one joke in one conversation were over 9% more likely to be voted into leadership positions. Bosses with any sense of humor are 27% more motivating and admired than those with none, and their employees are 15% more engaged.

The individual risk that one-off jokes will flop is much smaller than the aggregated risk of a lifetime of boring, disengaged conversation. But our worries aren’t totally unfounded—the trick is getting it right.

The Benign Violation Theory

Psychologists Peter McGraw and Caleb Warren suggest humor works when things are neither too benign (boring) nor too violating (scary, aggressive, inappropriate)—but somewhere between.

Simply walking downstairs: benign. Falling and breaking your arm: violation. Pretending to fall, landing in a silly pose: benign violation.

The challenge: the sweet spot moves depending on context, relationship, and individual humor styles. What seems funny from sweet-Olivia might seem mundane from sarcastic-Jimmy.

Affiliative vs Aggressive Humor

Affiliative humor brings people together, makes everyone feel included, increases psychological safety.

Aggressive humor involves put-downs or insults, decreases psychological safety, fragments people into categories.

When in doubt, be gentle. It’s less costly to flop for being too benign than too aggressive. Failed gentle humor means people don’t laugh and move on. Failed aggressive humor means hurt feelings that take a long time to heal.

Self-Deprecation: The Technical Professional’s Secret Weapon

Former congressman Ric Keller traces his political career start to a joke at age 34. After sitting through hours of serious speakers, he opened: “I feel like Elizabeth Taylor’s seventh husband on his wedding night. Technically, I know what I’m supposed to do. But at this point, I don’t know how to make it interesting.”

Cheesy? Sure. But it got a huge laugh from a crowd hungry for a fresh voice. Self-deprecation works best when leaders and high-status people use it—it makes them seem more approachable and human.

For technical professionals presenting to executives:

  • “This dashboard has so many charts I’m not sure if it’s a Tableau workbook or modern art”
  • “I’ve explained this architecture so many times I dream in flowcharts”
  • “Our data pipeline is held together with duct tape and prayers—but it’s documented duct tape and prayers”

Self-deprecation shows vulnerability while maintaining competence. You’re acknowledging imperfection without undermining your expertise.

What Levity Isn’t

Don’t make fun of stakeholders, team members, or anyone with lower status. “Punching down” kills psychological safety and trust. The New Yorker cartoon wisdom—“If you can’t say something nice, say something clever but devastating”—might work for comedy, but it’s terrible for professional relationships.

If you’re choosing between gentle-but-cheesy and clever-but-devastating, choose the former. Everyone likes to laugh, but nobody likes to be laughed at.

For Data Engineers: Practical Levity

Levity isn’t about being a comedian. It’s about energy management—keeping conversations in the engaged quadrant. Specific tactics:

In presentations: Start with a light observation. “I know the last thing anyone wants Monday morning is another architecture diagram, but I promise this one has colors.” Not hilarious, but it acknowledges shared reality and signals you’re human.

In meetings: When discussions get too heavy, inject perspective. “We’ve been debating this for 30 minutes. I propose we decide by seeing who can juggle the most coffee cups.” Gets a laugh, resets energy, helps people realize they’re overthinking.

In Slack: Use gentle humor to soften messages. Instead of “This code has bugs,” try “Found some unexpected features in the code—I think the bugs are multiplying via asexual reproduction.”

In documentation: “Note: This regex looks like someone sat on the keyboard. I promise it makes sense.”

The goal isn’t entertainment—it’s keeping people engaged when discussing complex topics. Small moments of levity prevent cognitive fatigue and make information more memorable.

How This Applies to LLM Communication

LLMs respond to tone. Compare:

No levity: “Explain dependency injection”

With levity: “Explain dependency injection like I’m a curious golden retriever who somehow learned to code but still gets confused by fancy patterns”

The second isn’t just more fun—it actually produces better results because it gives Claude clear framing about complexity level and explanation style. You’re using levity to establish shared context.



K Is for Kindness: Putting Others’ Conversational Needs First


Brooks’ final maxim is the most challenging: trying your utmost to put others’ conversational needs first. Not always possible, not always optimal, but reaching for it relentlessly gives you the best chance of being the partner you mean to be.

The Anderson Cooper and Stephen Colbert conversation about grief demonstrates this beautifully. Both men lost fathers and brothers tragically young. When Cooper interviews Colbert on CNN—no live audience, just two people in director’s chairs—they create one of the most powerful public conversations on grief you’ll watch.

What makes it work isn’t just the topics (grief, loss, mothers) or the questions (14 questions in 20 minutes) or even the levity (warm laughter every 2-3 minutes). It’s their invisible but palpable focus on the other person—their words, their struggle, their life, their needs.

Cooper prepared extensively—read previous Colbert interviews, can quote his words, has notes and questions ready. Colbert is self-deprecating, allows Cooper to lead, responds to vulnerability with vulnerability. When one offers a thought, the other builds on it or discloses in return.

Ten minutes in, Cooper says: “I wish I had, like, a scar. A Bond villain scar running down my eye and face” to signal “I’m not the person I was meant to be.”

Colbert shifts in his seat. Takes off his glasses. Smiles. “But you’re entirely the person you were meant to be.”

This is kindness in conversation—gently challenging a thought that might not serve Cooper, even when it creates tension. Cooper asks, voice breaking, “Do you really believe that?” Colbert pauses, maintains unwavering gaze, replies simply: “Yes.”

Kindness doesn’t mean agreeing with everything. It means figuring out what your partner needs and helping them get it—encouragement, hard feedback, new ideas, a laugh, a sounding board, challenging questions.

The Egocentrism Problem

Psychologist Jean Piaget showed that around age seven, children begin recognizing others have different perspectives. Before that: extreme egocentrism. Hands over eyes during hide-and-seek? They think you can’t see them. “That’s mine!” isn’t compelling justification—it’s self-centered assertion.

Here’s the devastating fact: we don’t shed egocentrism after age seven. It operates constantly in the background, sabotaging our ability to converse effectively.

Egocentrism makes us choose topics we like (assuming our partner will like them because we do). It makes us ask questions we find interesting rather than ones they’d be excited to answer. It makes us create humor that we think is funny without considering whether it serves them.

Researcher Boaz Keysar demonstrated this brilliantly. He had pairs sit back-to-back as speaker and listener. Speakers imagined scenarios (“suspect your friend has been planning a surprise”) then delivered lines (“What have you been up to?”) with any inflection. Listeners chose the intended meaning from four options.

Results: Listeners thought they correctly identified meaning 85% of the time. Speakers thought listeners understood 70% of the time. Actual accuracy: 44%.

Even more striking: Keysar repeated this with people who didn’t speak the same language. Chinese speakers, English listeners. American listeners believed they understood 65% of the time. Chinese speakers believed they’d been understood 50% of the time. Actual accuracy: 32%.

George Bernard Shaw: “The single biggest problem with communication is the illusion that it has taken place.”

We can’t prioritize others’ conversational needs if we don’t understand them, especially when we mistakenly believe we do. Kindness requires continuously checking that illusion.

For Data Engineers: Practical Kindness

In technical discussions, kindness means matching your partner’s level:

With junior engineers: Don’t overwhelm with advanced concepts. Ask “Does this make sense?” and actually wait for response. Explain the “why” behind decisions, not just the “what.”

With senior leaders: Don’t bury them in technical details. Start with business impact, then offer technical depth as needed. Ask “What level of detail would be helpful?”

With non-technical stakeholders: Use their domain language, not yours. Instead of “We need to denormalize the schema to improve query performance,” try “We need to restructure how we store data so reports run faster. Think of it like rearranging a library—same books, faster to find what you need.”

In difficult conversations, kindness means creating safety:

When delivering bad news: “I need to tell you something that’s going to be disappointing” (prepares them) vs. abrupt revelation.

When receiving feedback: “I appreciate you bringing this up—help me understand what you’re seeing” vs. defensive response.

When disagreeing: “I see it differently—can I explain my thinking?” vs. “That’s wrong.”

In team meetings, kindness means ensuring everyone’s heard:

Notice who hasn’t spoken: “Jamie, you’ve been quiet—what’s your take on this?”

Build on others’ ideas: “Building on what Alex said…” vs. introducing competing idea.

Redirect credit: “That was Sarah’s insight—Sarah, can you explain how you approached it?”

The Hardest Part

Always putting others’ needs first is unrealistic—and sometimes not optimal. You have legitimate needs on the low-relational side of the conversational compass. The advice isn’t to be selfless martyr.

The advice is to try relentlessly. To continuously ask yourself: “What does this person need from this conversation?” Then help them get it.

Topics and questions provide substance. Levity keeps people engaged. Kindness creates space where meaningful exchange can flourish—where people feel respected and valued. It’s the master class that takes other skills to elite level.

How This Applies to LLM Communication

Yes, kindness applies to AI conversation. Not because LLMs have feelings, but because kindness frameworks produce better results.

Provide context about what you need:

Bad: “Write me a function”

Good: “I need a function that parses JSON from an API response. I’m fairly comfortable with Python but new to this API, so I’d appreciate comments explaining the approach.”

Acknowledge what’s working:

“That solution worked perfectly for the first case, but when I tried it with nested JSON, I got an error. Here’s what I’m seeing…”

Be specific about confusion:

Instead of “This doesn’t make sense,” try “I understand the first two steps, but I’m confused about why you’re using a set instead of a list in step 3. Can you explain the reasoning?”

You’re not being polite to a machine—you’re practicing communication patterns that work with humans too. Every interaction is practice for the coordination game.



The Through-Line: System Design for Human Connection


Brooks’ research points to a fundamental reframe: “The key to good conversation isn’t knowing, but learning. It’s about being interested in your partner, not interesting yourself.”

This should resonate with analytical professionals. You don’t need charisma or extroversion. You need to deploy genuine curiosity—something most technical people already possess in abundance for technical problems—toward the humans you work with.

The TALK framework provides structure:

T - Topics: Prepare 3-5 topics before important conversations. System 2 thinking frees mental space for system 1 responsiveness.

A - Asking: Ask more follow-up questions. They signal listening, care, and thoughtfulness. Master conversationalists don’t need specialized expertise—they have specialized curiosity.

L - Levity: Manage energy, not entertainment. Small moments of playfulness prevent disengagement and make information memorable.

K - Kindness: Continuously try to understand and meet others’ conversational needs. Check your egocentrism at the door.

These aren’t difficult techniques: brainstorm topics beforehand, ask more follow-up questions, inject occasional levity, focus on your partner rather than yourself. What makes them powerful is consistent application.

The Career Impact

Communication skills ranked #2 among the most important factors when hiring engineering project managers—higher than technical proficiency at #10. Executive presence accounts for 26% of what it takes to get promoted. Google found their most productive employees weren’t those with highest technical skills but those with developed soft skills who could turn expertise into influence.

Research shows conversation is a learnable skill with predictable patterns. For data engineers willing to treat communication with the same analytical rigor we bring to technical problems, the evidence suggests significant returns—career advancement, team performance, stakeholder trust.

You don’t need to become someone different. You need to become more strategically curious. The research shows exactly how.

Implementation Checklist

Before your next stakeholder meeting:

  • Brainstorm 3-5 questions based on their likely concerns
  • Prepare topics that climb beyond status updates
  • Plan one moment of levity (even just acknowledging shared reality)
  • Identify what they need from the conversation

During the conversation:

  • Ask at least one follow-up question for every major point
  • Notice when you’re shift-responding vs support-responding
  • Watch for lower-left quadrant (disengagement) and inject energy
  • Check: am I making this about me or about them?

After the conversation:

  • What topics worked? Which fell flat?
  • Did I ask enough questions?
  • Was there authentic laughter?
  • Did they seem to get what they needed?

Start with one conversation. Apply one element of TALK. Notice the difference. Build from there.

The same framework that makes you better at technical writing makes you better at stakeholder management. The same curiosity that drives your debugging makes you better at understanding people. You already have the skills—you just need to deploy them systematically.

Further Reading


Brooks, Alison Wood. Talk: The Science of Conversation and the Art of Being Ourselves. Random House, 2024.

Key research papers from Brooks and colleagues:

  • Brooks, A. W., & John, L. K. (2018). “The Surprising Power of Questions.” Harvard Business Review. (On question-asking in sales, dates, and negotiations)
  • Huang, K., Yeomans, M., Brooks, A. W., Minson, J., & Gino, F. (2017). “It Doesn’t Hurt to Ask: Question-Asking Increases Liking.” Journal of Personality and Social Psychology. (Speed dating and question types study)
  • Brooks, A. W. (2014). “Get Excited: Reappraising Pre-Performance Anxiety as Excitement.” Journal of Experimental Psychology: General. (Anxiety reframing research)

Related conversation science research:

  • Derber, C. (1979). The Pursuit of Attention: Power and Ego in Everyday Life. Oxford University Press. (Conversational narcissism and shift vs support responses)
  • Edmondson, A. C. (2018). The Fearless Organization: Creating Psychological Safety in the Workplace for Learning, Innovation, and Growth. Wiley. (Psychological safety research)

For technical professionals specifically: