Company-level AI bets should anticipate improving model capabilities rather than being limited by today's performance. Because AI systems require experimentation and iteration, the speakers argue that transformation programs should operate like growth projects instead of conventional software deployments.

Choose future-important use cases before models fully mature
Build infrastructure in anticipation of capability gains
Assign a substantial pod to major company-level bets
Use iterative growth methods rather than one-time IT deployment

“The model capability, don't worry that it cannot do what you want to do today. Build the infrastructure and the setup to do that thing,…”
Kieran Flanagan · 18:00

“The big thing that I think and what I've seen in other companies is it should be run like a growth project, not an IT…”
Kieran Flanagan · 18:00

#ai strategy#growth#transformation#operations

Explainer· 4

Explainer02:00

Where Kraken First Applied AI Across Marketing

Kraken began experimenting with AI in creative production, performance marketing, translation, and customer research. Performance work benefited from rapid variation, while the global organization used translation across product and off-product experiences.

Creative and performance marketing were early adoption areas
Performance campaigns benefit from greater asset velocity and variety
Translation supports Kraken's multilingual European footprint
AI accelerates qualitative and quantitative research

“And to be honest, the two areas where we started using AI and experimenting was around creative, both from a brand and storytelling standpoint as…”
Mayur Gupta · 02:00

“So, you know, translation has been a huge area for us, both in product and off-product.”
Mayur Gupta · 02:30

#ai adoption#marketing#creative#translation#research

Explainer05:00

AI Can Provide Fast, Synthetic Customer Feedback

HubSpot uses model projects loaded with buyer-persona information to review product pages and brand campaigns. This creates a fast feedback loop during inception rather than waiting for a completed asset and a conventional research study.

Load buyer-persona material into a dedicated AI project
Ask the simulated customer what resonates and what does not
Gather feedback while an asset is being conceived
Reduce dependencies between growth and research teams

“And so anytime we're writing a product page or building a brand campaign, doesn't matter what it is, we can just ask basically a fictional…”
Kipp Bodnar · 05:00

“So it's also helps them get a deeper understanding of what the customers may respond to, and also bringing more agility because you are now…”
Mayur Gupta · 06:00

#customer research#personas#feedback#campaigns

Explainer08:30

The Unstructured-Data Unlock Inside a Company

The speakers describe consolidating PDFs, decks, documents, meeting recordings, and transcripts into repositories that employees can query. Kraken uses a local instance for sensitive information, while HubSpot documents workflows and records meetings to create reusable organizational context.

Centralize scattered documents and transcripts
Make accumulated knowledge directly queryable by employees
Document workflows before deciding what to automate
Use local instances when confidential data must remain controlled

“So one of the things that we are doing at Kraken is we've created our local instances where all the unstructured data is actually being…”
Mayur Gupta · 09:00

“And then the second thing is every meeting is recorded.”
Kipp Bodnar · 10:30

#unstructured data#knowledge management#documentation#privacy#automation

Explainer11:30

When to Buy AI Seats, Build on an API, or Run Locally

The discussion separates general experimentation from focused production use cases. Business seats are a practical starting point, API applications can be dramatically cheaper for narrow workflows, and locally hosted open-source models suit specific tasks involving privacy-sensitive data.

Buy frontier-model seats for broad exploration
Build on APIs for focused and repeatable use cases
Consider local open-source models for privacy-sensitive workloads
Evaluate whether the capability is core enough to justify owning the IP
Accept an 80% solution when customization costs exceed its benefit

“I would still always want to start with off-the-shelf because I would want to try to prove the use case as quick as possible versus…”
Kieran Flanagan · 12:30

“If you are general use cases trying to figure that out, I think probably going and buying those seats is probably not a bad place…”
Kipp Bodnar · 15:30

#build vs buy#saas#api#open source#data privacy

Story· 1

Story24:00

Mayur Used Claude to Diagnose Household Overspending

Mayur describes cleaning personally identifiable information from household financial data and analyzing CSV files with several AI models. The exercise corrected assumptions about Amazon spending and exposed unexpectedly high Uber costs, reinforcing his belief that professional AI fluency starts with personal experimentation.

Remove personally identifiable information before uploading data
Use real personal problems to build practical AI fluency
Interrogate transaction data instead of relying on assumptions
Carry lessons from personal experiments into professional workflows

“So what I did was I cleaned up all my data, I removed all the PI, man, I dumped all the CSV files into Claude…”
Mayur Gupta · 24:30

“And then we so I think the only way you unlock AI in your professional life is when you're actually living and breathing it and…”
Mayur Gupta · 25:00

#personal finance#ai fluency#claude#data analysis#privacy

Tool· 3

Tool27:00

Turning Company Data Into an AI-Generated Growth Plan

Kieran demonstrates using a reasoning model to identify priority metrics and propose experiments for a synthetic SaaS business. The conversation explains that results improve when real operating data, previous experiments, customer journeys, UX flows, and external market signals are added as context.

Ask the model to identify the few metrics most relevant to a target
Turn those metrics into hypotheses and experiments
Load prior experiments to avoid repeating failed ideas
Add customer journeys and UX flows to ground recommendations
Correlate anonymized internal trends with external geographic signals

“But actually, here's all of the experiments we've run. So you have a library of all previous experiments, which again speaks to the fact the…”
Kieran Flanagan · 30:30

“And I suspect you're gonna get a first version of a growth plan that needs to be edited, but does not in any way need…”
Kieran Flanagan · 31:00

#growth strategy#reasoning models#experimentation#analytics#context

Tool35:00

Attach Each AI Assistant to One Defined Project

A project-specific assistant can ingest updates, experiments, meeting transcripts, and other artifacts from one folder without demanding elaborate structure. It can then retrieve historical experiments, explain outcomes, track follow-ups, and surface dependencies that teams otherwise struggle to reconstruct.

Define one assistant around one project or measurable goal
Put every relevant project artifact into one shared folder
Query past experiments and their outcomes conversationally
Turn meeting transcripts into maintained follow-up tables
Use accumulated history to preserve learning across team changes

“The key is to attach them to a singular project.”
Kieran Flanagan · 37:00

“So it's not only that we are inefficient because it's it's so manual, but then we become ineffective because we are not technically applying the…”
Mayur Gupta · 37:30

#project assistants#knowledge management#experiments#meetings#productivity

Tool39:00

Claude for Writing, OpenAI for Strategy, Gemini for Google Context

The hosts offer a task-oriented snapshot of model strengths: Claude for writing and creative work, OpenAI's reasoning models for strategy, and Gemini when Google Drive integration provides valuable context. They also warn that endlessly switching models may matter less than developing deep fluency with one capable system.

Use Claude for writing and creative marketing work
Use OpenAI reasoning models for strategic analysis
Use Gemini when Google Workspace context is decisive
Focus on two or three major platforms before exploring niche tools
Prefer deep adoption fluency over constant model switching

“Claude is a better creative model than OpenAI and Google.”
Kieran Flanagan · 39:30

“I think the right thing to say is all of these things are transformatively powerful and we are underusing them.”
Kipp Bodnar · 42:30

#claude#openai#gemini#model selection#ai fluency

Takeaway· 2

Takeaway03:30

Why Kraken Uses AI for Brand Ideas, Not Final Product Visuals

Kraken finds AI more useful for generating brand concepts and variations than for producing finished assets. Product-focused campaigns require accurate interfaces, graphs, and dashboards, where conventional production remains more dependable.

Use AI to expand the range of creative ideas
Treat generated variations as inputs rather than finished work
Use traditional production when product-interface accuracy matters
Match the production method to the asset's precision requirements

“And in brand, I think it always is where AI is helping us get smarter and uh and come up with more ideas, idea generation…”
Mayur Gupta · 03:30

“So what we are learning is, you know, it's great to give us ideas, it's great to give us different variations, but when we are…”
Mayur Gupta · 04:00

#brand#creative#product marketing#ai limitations

Takeaway18:30

Effective AI Adoption Needs Both Local Owners and Specialists

The recommended operating model combines business teams that own use cases with centralized AI and automation specialists. Domain teams identify valuable workflows and assemble relevant data, while specialists handle technically demanding automation and custom systems.

Let business teams originate use cases and desired outcomes
Make domain teams responsible for gathering relevant context
Use specialists for complex automation and custom software
Provide every employee with tools and permission to experiment
Treat curiosity and self-directed learning as individual responsibilities

“I think you have to have both of what Kieran just said is the honest answer.”
Kipp Bodnar · 20:00

“Anytime they are the ones generating the inertia, they are the ones pushing the use case.”
Mayur Gupta · 21:30

#operating model#enablement#teams#automation#leadership

Frameworks in this episode

AI Build-Buy-Local Decision Framework

Contextualized Growth Intelligence Loop

Depth-First AI Fluency Rule

Documentation-First AI Automation

Hybrid Domain-and-AI Ownership Model

Iterative Reasoning Stack-Rank

Single-Project AI Assistant

Synthetic Customer Feedback Loop

Three-Layer AI Adoption Model

Insights & moments

◆Hot Take· 1

Run AI Transformation Like Growth, Not IT

✶Explainer· 4