Promptly Speaking

On this episode of Promptly Speaking, Dan and I sit down with Diego Delgado, Senior Director of Strategy and Ops at Visa, to talk about how AI is actually used inside a company of 40,000 people. As an ai podcast focused on practical, real workplace stories, we get into why Diego treats AI as a thought partner rather than an answer machine, what he calls the "AI grenade," and why thinking before prompting matters more than ever for ai leadership at scale.

Key takeaways:

• Diego uses AI to structure his thinking, not to generate the answer itself
• The "AI grenade" is content that looks polished but explodes when nobody understands how it was built
• Critical thinking has to happen before you prompt, not after
• New employees need coaching on how to use AI responsibly, not just encouragement to adopt it
• Using AI doesn't remove your ownership of the final product
• Diego's background in consulting shaped how he breaks down problems before involving AI
• Good leadership means coaching people through mistakes, including AI-related ones

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 Cold open on structuring your thinking

01:08 Welcome and introducing Diego Delgado

01:27 Diego's path from Lima to BCG to Visa

04:37 Using AI as a digital thought partner

09:06 Why polished AI answers can be deceiving

10:23 Real examples from a 40,000 person company

15:34 Where the term AI grenade came from

18:06 The Clay founder's rules for sharing AI work

20:00 Thinking before prompting

26:06 Why cloth cannot be a shield for ownership

27:55 Personifying AI as a teammate

32:52 Building critical thinking without a consulting background

41:00 Diego's closing advice

Nobody files an insurance claim because something good happened. That's the honest, slightly strange truth at the center of this episode.

I'm joined by Aaron Davidson, a 22 year insurance industry veteran who has worked everywhere from McKinsey to Guidewire to ServiceNow to his own insurtech startup, and who spent the last year going deep into building with AI coding tools. We talk about what AI is actually good at (remembering, finding, automating repetitive tasks) and where it still falls short (judgment, empathy, weighing outcomes over time). Aaron explains why insurance is a probabilities business, not a guarantees business, why that makes it uniquely hard for AI to fully take over, and how he's using what he calls "empathetic product design" to build tools, including a daily AI news briefing, based on problems he lived himself. We also get into why roles are changing faster than job titles can keep up, and why some trades are nowhere near replaceable.

Key takeaways:

• Insurance runs on probabilities, not guarantees, which is why judgment still matters
• AI is strong at memory, retrieval, and repetitive tasks, weak at empathy and weighing outcomes
• Claims adjusters can handle more active files when AI handles summarization and recall
• "Empathetic product design" means building tools for problems you've personally lived
• Vocational trades like construction and auto mechanics are far from AI replacement
• Problem solving skills matter more than technical skills as building gets easier
• Historical tech shifts (steam engine, railroad, internet) took decades to fully land

Connect with Aaron Davidson: https://www.linkedin.com/in/aarondavidson1/

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

Timestamps:

00:00 Cold open: computers and judgment

01:00 Welcome and introducing Aaron Davidson

01:18 Aaron's 22 years in insurance

05:54 Moving into tech at Guidewire

08:18 What is insurance, really

10:09 Where AI takes the insurance industry

12:01 Case summarization in claims

13:31 Why AI struggles with judgment

15:18 What actually bugs Aaron about insurance

20:58 Do we need to do it this way at all

27:04 Could a neighborhood run its own insurance

33:09 Building the Newsletter AI tool

36:50 Empathetic product design explained

44:52 AI as the next industrial revolution

51:08 The limits of vibe coding

53:37 Why being human still matters

53:39 Closing thoughts

Six months ago we were told to use AI constantly, for everything. Now companies are capping it, and we have to relearn how to work. This episode of our ai podcast looks at the cognitive overload that comes with juggling AI across five workflows at once, why output can rise while understanding quietly drops, and what that means for the future of work.

Key takeaways:

• Model providers were subsidizing token costs, and usage-based pricing is changing how companies let employees use AI
• Managing multiple AI-assisted workflows at once creates a different, heavier kind of context switching
• Output can go up while depth and understanding quietly go down
• Executives and knowledge workers alike are skipping the deep internalization step AI makes easy to skip
• Speed and quality are in real tension, and planning time before a task may matter more than the task itself
• The fix may be small, deliberate, AI-free planning time before diving in

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 Cold open: retraining after AI limits

00:41 Welcome back to the studio

01:20 Six months of "just use it"

02:03 How company token caps happen

02:48 Models got better, usage exploded

03:35 A friend's company caps his AI use

04:34 Companies rolling back heavy spend

04:58 Choosing the right model, not just the biggest

05:38 Running ten tasks at once

06:33 Context switching gets harder

08:15 Juggling searches, Slack, and tools

10:09 New research on context switching and output

11:46 Cognitive load and exhausted brains

12:20 Prepping executives before AI

13:46 Shallow understanding as a side effect

14:21 Balancing speed and quality

15:23 More capacity just means more expected of you

17:52 What happens when your AI gets cut in half

18:43 A simple fix: plan before you prompt

Universities are basically small cities, and getting everyone to agree on AI is nearly impossible. Beth Myers, CIO at the University of Colorado Denver, joins Sara and Dan to talk about rolling out AI across a massive, opinionated institution. She covers the heated debates between faculty and students, why some disciplines are all in while others resist, what the early research says about AI and learning, the mental health concerns showing up on campus, and why she still worries about the environmental cost of it all.

Connect with Beth Myers: https://www.linkedin.com/in/beth-myers/

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 Cold open, AI and complex organizations

00:54 Beth's title and 27 years at CU

01:31 What a vice chancellor actually does

02:11 Why bringing AI to a university is complex

04:11 Losing the ability to have hard conversations

04:31 Who gravitates toward AI, who resists

05:43 Talking to the AI student society

06:36 The computer lab techs' hesitancy

07:23 How heated does it get

09:21 How AI is changing how people learn

11:32 Why students need clearer AI guidelines

11:54 AI and entrepreneurship in students

14:13 The risks of vibe coding without knowing security

15:47 AI literacy and parenting teenagers

17:15 The parental controls Beth actually uses

19:56 AI, loneliness, and student mental health

21:08 Bringing AI into a university's shared governance

23:08 Why the ChatGPT rollout didn't go smoothly

24:10 Required training before getting AI access

26:14 Students who don't want AI in their education

27:22 What early research says about learning and AI

33:39 The environmental cost of AI

35:52 AI subsidies and what happens when prices rise

41:26 Closing thoughts and thanks

A fluent AI answer isn't the same thing as a correct one, and that gap is costing companies more than they realize. Andrew Marks, co-founder of Success Hacker, joins Sara and Dan to unpack why AI's confident tone tricks even smart people into skipping the fact-check. They dig into persona validation, confidence calibration, and why Andrew's team built their own small language model instead of leaning on a chatbot. Along the way: why customer success teams are selling every day whether they admit it or not, why "easy button" layoffs are a mistake, and how to actually put a number on value.

Key Takeaways:

• AI sounds confident even when it's wrong, and that fluency is exactly what makes it dangerous to trust blindly
• Persona validation (feeding AI a stakeholder's real background) surfaces angles a person might miss
• Confidence calibration forces AI to rate its own certainty, which helps catch hallucination before it spreads
• Customer success teams often assume customers understand the value they're getting, when they don't
• Value looks different depending on who you're talking to: end user, middle manager, or C-suite
• Customer success professionals are selling constantly (renewals, expansions, retention) even when they resist the label
• Cutting customer-facing teams because AI can "do it all" is a mistake companies will regret
• Small, domain-specific language models may end up mattering more than giant general-purpose ones
• Real value has to be translated into numbers stakeholders care about, like budget or revenue impact

00:00 Cold open: the easy button myth

00:43 Welcome and introducing Andrew Marks

01:10 Success Hacker's accreditation and university ties

02:11 What customer success actually is

04:34 Defining value across stakeholders

06:23 The biggest AI mistake teams make

07:19 AI hallucinations in the real world

08:23 What persona validation really means

09:08 Using AI on a real RFP response

13:27 A better way to prompt and iterate

17:32 Building a knowledge graph and trust layer

19:16 Four years building an AI platform

21:25 Who actually owns customer value

23:43 How this reshapes the customer journey

25:11 Naming "bot shitting"

26:10 Why layoffs based on AI are a mistake

27:58 AI should work with you, not for you

29:28 Teaching critical thinking in the AI era

32:47 Small language models vs. large ones

36:39 What conventional CS wisdom got wrong

39:10 The $150 million case for selling

42:04 Why comp models haven't caught up

44:41 Making value measurable and portable

48:35 The Project Lead the Way story

50:51 Advice for ethical AI adoption

53:52 What to do with 30 minutes and AI

References & Links:

• Project Lead the Way, STEM curriculum nonprofit (www.pltw.org)
• Success Hacker's AI for Customer Success training program (www.successcoaching.co/ai-for-customer-success)
• University of San Francisco School of Executive Management (www.usfca.edu/management/programs/executive-education)

He helped rebrand Mark Cuban's network, managed bands in the 90s, and produced the Jackass guys.

Now Tim Glomb spends his days as an AI transformationist, and he's not selling hype.

In this episode, Tim joins Sara and Dan to talk about what actually happens when a company adopts AI seriously (spoiler: it's not "slap it on top"), why the open web's foundation is quietly crumbling, and how a small bar owner turned empty seats into free beer and $200 nights using a simple text campaign. We also get into his dad's tear-jerking ChatGPT moment, AI-generated music, and why he worries more about social media than about any AI model.

Key takeaways:

• Most companies are putting "AI paint" on broken processes instead of rebuilding the foundation
• The open-source code holding up the entire internet is maintained by a shrinking handful of people
• Small, local businesses can scale further and faster than ever using AI plus real business acumen
• A simple photo-and-text workflow turned empty bar seats into real revenue
• Context is everything when prompting AI, vague questions get vague, even misleading answers
• Tim's dad had an emotional breakthrough using ChatGPT to redesign his bathroom
• Suno.com lets anyone turn a memory or a person into an original song in minutes
• Nuclear energy, especially microfusion, may be the real answer to AI's energy demands
• Some questions (therapy, sensitive personal decisions) simply aren't ones you should hand to AI alone

00:00 Cold open

00:46 Meet Tim Glomb

01:50 The biggest AI misconception execs have

03:15 "AI paint" on a junk car

04:31 Where human glue lives in a business

07:16 Becoming the operator, not getting replaced

09:02 Code review bottlenecks from AI-written code

10:48 The internet's crumbling open-source foundation

13:27 His dad's ChatGPT bathroom breakthrough

15:50 A resurgence for local economies

18:43 One small business, one big AI use case

21:02 Turning a bar's empty seats into $200

25:15 Advice for kids choosing a career path

29:14 The therapy question you shouldn't ask AI

32:10 Why context changes everything you ask AI

36:03 Scanning food with the Yuka app

38:09 Estrogenic compounds, wastewater, and AI research

40:04 AI's real climate cost

41:07 Betting on nuclear fusion energy

44:08 Making AI-generated songs with Suno

48:09 Why he blames the music industry, not AI

Connect with Tim Glomb: https://www.linkedin.com/in/timglomb/

Follow Sara: https://www.linkedin.com/in/saralynneroberts/

Follow Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

AI can build almost anything now, so what's actually left for a product manager to do? Cliff Barrett, who has scaled product teams at Mailchimp, ChowNow, and Iterable, joins Sara and Dan to unpack why judgment, not tooling, is becoming the real differentiator in product leadership. We talk about a CNN test that 20 million users ignored, why "bring a prototype, not a presentation" is the new standard, the fatigue PMs feel from constant AI-powered output, and how Cliff built his own personal accountability agent to track his fitness journey. It's a grounded, practical look at leading through the AI shift.

Key takeaways:

• Judgment, not data access, is what separates a good product decision from an obvious one
• A CNN customization feature that tested well in feedback flopped with 20 million real users
• "You don't bring a presentation anymore, you bring a prototype" is becoming the new baseline
• PMs are experiencing a new kind of fatigue: constant bombardment with fast AI-built options
• Curiosity, more than credentials, is what Cliff screens for when hiring product people
• Demo days and hackathons help channel company-wide building without losing focus
• Cliff built his own AI accountability agent to track workouts, meditation, and eating patterns
• Stepping away from screens, and away from AI itself, is still essential to good thinking

00:00 Cold open: judgment over AI hype

00:50 Meet Cliff Barrett

01:05 What a product manager actually does

01:51 The Airbnb product management debate

03:07 Biggest risk: trusting AI's "right" answer

03:26 The CNN newsfeed test that flopped

04:42 How Iterable used AI for marketers

05:45 Why you still need to ask "why"

06:50 Market analysis, data, and stakeholder input

07:25 PM fatigue and decision overload

08:01 The case for thinking time and naps

09:33 Skills for the next generation of PMs

10:22 The Nokia to Apple mobile shift

11:02 Prototypes over presentations

11:27 Building HTML sites at Glean

12:00 Everyone adopts AI at a different pace

13:29 Discernment: what actually matters in a demo

14:16 Step one for product leaders right now

15:19 When everyone can build, how do you focus

16:20 Running successful demo days

18:10 What makes Cliff want to join a company

19:36 Agents marketing to agents

21:08 Why product managers still matter

22:00 Cliff's personal AI accountability agent

24:01 What the historical look-back actually reveals

26:34 How Cliff screens for curiosity in hiring

28:40 Avoiding burnout as a builder

31:00 Using AI to simplify information overload

32:03 AI as a thinking partner, not therapy

34:07 Final advice for product leaders

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

Can AI make kids better learners, or just more dependent? An educator's honest answer.

For our first ever in-person guest, we sat down with Larame Spence, who has spent years tutoring, running a private school, and building AI learning tools. His take flips the usual fear: the most powerful use of AI in education right now isn't sitting a kid in front of a chatbot, it's using AI as the source a parent, tutor, or teacher works through with the student, so the human stays in the loop.

Key takeaways:

• Why AI works best as the tutor's source material, not the tutor
• The MIT study where feedback labeled "human" beat identical feedback labeled "AI"
• Building a whole curriculum around one kid's obsession (the all-golf student)
• The calculator question, and what we lose and gain with every new tool
• "Confidently wrong" AI, and teaching kids to check their resources
• How a vetted library plus retrieval makes AI answers more trustworthy
• The privacy trade-off of an AI that learns exactly how your child thinks
• Why struggle, not ease, is where the real learning lives

This one will change how you think about AI and your kids.

Connect with Larame Spence: https://www.linkedin.com/in/laramespence/

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 The scary parts of offloading to AI

01:30 Our first in-person guest

02:30 How education is going to change

03:30 AI as the source, not the tutor

05:30 Tutoring Algebra 2 with an AI

07:30 The MIT human vs AI feedback study

10:30 Spotting AI-generated content

12:00 Helping kids across a language barrier

14:00 The Dominican Republic email workflow

16:30 Will we still learn other languages?

18:30 Tailoring the lesson to each student

21:30 Asking questions without feeling dumb

24:00 Rethinking how we assess students

26:00 The calculator question

29:00 The brain is a muscle

31:30 Advice for nervous parents

33:00 The all-golf student

36:00 Personalizing a lesson by interest

38:30 Measuring student confidence

40:30 Confidently wrong: the star ceiling

43:30 Teaching kids to check their resources

46:00 Cutting hallucinations: tools and retrieval

49:30 How the teacher's role changes

53:00 Does AI help long-term recall?

55:30 Why they built the interactives first

58:30 The privacy trade-off for kids

60:30 Higher ed and the path to a job

63:30 The God code and earned struggle

66:30 Brandon Sanderson and the experience of a book

69:00 Sycophantic AI and why struggle matters

73:00 One takeaway, and don't underestimate Nano Banana

Most people are still typing questions into a chatbot. Brian Albert is building software while he sleeps.

Brian went from running a digital marketing agency with a team of content creators to replacing that whole function with AI the month ChatGPT arrived, and growing a YouTube following of 25,000 by simply giving away what he figured out. In this conversation we get tactical: the email workflow he runs for about $30 a month, why he builds his own tools instead of renting someone else's wrapper, what "disposable software" means, when to use a project versus a gem, and the one thing he thinks actually lasts. Playfulness.

Key takeaways:

• Why Brian replaced content creators with AI in January 2023
• The email-triage workflow anyone can build for ~$30/month
• Build vs. buy: the case for owning your own automation
• "Disposable software": agents that build throwaway tools on demand
• Projects vs. gems vs. custom GPTs, and when each one fits
• How to get unstuck by asking a different model
• The skill Brian says will still matter in six months

Connect with Brian Albert: https://www.linkedin.com/in/brian-albert-executive/

Follow Sara & Dan:

Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 You can do anything with AI

00:47 Only 0.04% automate with AI

02:46 How Brian feels about this moment

04:07 Firing the content team in 2023

05:04 Six friends, then 25,000 followers

06:17 Giving everything away for free

07:39 A first agentic email workflow

09:42 Is it actually hard to build?

12:39 The real monthly cost, about $30

16:54 Build it yourself vs. buy the product

19:48 Add the next step to the automation

22:11 Was Brian always technical?

22:58 Starting with a PRD

25:29 A voice-driven book-writing app

26:02 Getting unstuck by asking another model

27:56 Disposable software, explained

30:45 Building apps vs. a whole eng team

30:54 Cowork and scheduled agents

36:01 Is this feeding the AI slop problem?

38:54 Unique perspective, value, adjust

41:22 Spotting AI writing, the do-not-use list

43:01 One thing to go do right now

45:07 Projects vs. gems vs. custom GPTs

47:25 The skill worth investing in, playfulness

50:36 Wrap up

What if AI could make your employees feel more valued, not less?

Jason Stover, Executive Vice President at SSA Group and the builder behind Scout — SSA's internal AI platform — joins Sara and Dan to share how a people-first philosophy is driving one of the most thoughtful AI adoption strategies in the hospitality industry. From getting laid off teams through COVID, to discovering ChatGPT and not sleeping for weeks, to accidentally sending an AI-written email that almost got him flipped off by a colleague, Jason's story is equal parts honest, funny, and genuinely inspiring.

Key takeaways:

• Why the hospitality industry may be more AI-resistant than most — and why that's a strategic advantage
• How Scout was born, and what makes it different from generic AI tools
• The fight-flight-freeze response to AI adoption, and how to move your team past it
• Why AI adoption strategy starts with one question: "How does this impact me?"
• The email incident that changed how Jason thinks about AI communication forever
• How SSA is using behavioral psychology to personalize the AI experience for each employee
• What it really means to lead with empathy when rolling out new technology
• Why giving a long-tenured employee a four-day week might be the best ROI AI can deliver
• Jason's honest take on the future of work and the industries most at risk
• The number one piece of advice for any business leader who is still dancing around the AI topic

Connect with Jason Stover: https://www.linkedin.com/in/jason-stover-4b6b7a1b/

Follow Sara & Dan: Sara: https://www.linkedin.com/in/saralynneroberts/

Dan: https://www.linkedin.com/in/danroberts27/

Email the show: [email protected]

00:00 Episode preview & intro

01:15 Jason's parallel path: people & tech

01:43 Digital scales & predictive analytics

03:11 How Jason stays curious about AI

03:51 The YouTube channels he follows

04:44 EVP blocking Fridays for AI discovery

04:54 Journey from dishwasher to VP at SSA

06:38 The pivot: stepping down to focus on AI

06:47 COVID, layoffs, and the ChatGPT moment

09:47 Why it's always about people, not just AI

10:21 Meet Scout — SSA's internal AI platform

10:56 Tank Stover: the persona that started it all

14:08 Early feedback: fight, flight, or freeze

15:33 How employees are actually using Scout

17:26 Building career paths with AI in hospitality

18:10 Building trust in AI tools with frontline staff

18:47 The million-dollar question: transparency

19:55 What most CIOs get wrong about rollout

20:46 Trips and falls along the way

20:57 The email incident that changed everything

24:17 Rethinking who to bring AI to first

26:05 Repurposing saved hours for human connection

27:14 AI helping us be more human

27:41 Tangible vs. soft ROI of AI investment

28:47 The four-day workweek conversation

31:59 How Scout connects siloed departments

33:21 Crystal ball: the future of SSA with AI

36:15 Demographics, retention & daily pay

39:10 Leading with empathy — advice for leaders

39:31 Stop freezing: come up with a strategy

40:39 What Jason admires about Anthropic & Claude

41:13 A united leadership team behind the vision

42:16 Advice for young people navigating AI

43:04 Behavioral Essentials & personalizing Scout

46:20 The data trust problem

46:40 Using AI to discover your own strengths

47:18 Where to find SSA & closing thoughts

47:41 Outro