The AI First Opportunity for Scale-ups
Brandon: Before we get to Charlie, the first question I wanted to ask you - what is the opportunity right now for scale-ups to go hard on being AI-first?
Bethany: I made a face when you asked that question because I was wondering if you meant AI-first in the product or AI-first within the company. I think there's the bigger existential question: is your product shit and do you need to tear it up and just start it again as an AI product? And if you're more on the startup, the scale-up side, I would highly encourage you to do that.
Although it's interesting because everything is so new, there's a lot of scope for improved products. Like for me, the really big gap is the next generation Zapier - something that is really easy for non-technical users to use to automate the soul-destroying work in their lives.
There are loads and loads of companies trying to figure out agentic AI, but they're still technical for technical people rather than no-code, easy automation for non-techies. Once that unlocks, then everything in your business should be AI-first.
It's a bit of a question of AI agents, which are going to be verticalized right now for very specific functions as opposed to AI agents working together in tandem to actually have more of a net effect for an organization.
Bethany: My nirvana, and I think at some point this year will be: I have all of these CVs to sort through. I don't want to sort through them. I tell the agent, these are the topics that I'm looking for on the CVs. This is what good looks like. Go and read all of these and bring out the ones that I'm looking for rather than a rules-based system.
Implementing AI in Your Startup: Practical Steps
Creating AI Policies and Infrastructure
Brandon: When you think about rolling out AI within the organization right now in 2025 for a company, what makes sense? Would you encourage companies to create a policy, license some tools, and launch them into the company?
Bethany: All the technology is changing so rapidly that I would not commit to a single one. I would go with rolling monthly contracts and continue to experiment. It's too early to sign up for annual commitments to anything right now. Things are moving too fast.
Policies definitely, because it makes you think through all of the gotchas and it makes it really clear for the organization.
Reward Mechanisms and Team Engagement
At Peak, we have a few different ways that we are focusing on the team using and getting more experimental with Gen AI:
Weekly Show and Tell: In our company all-hands, there's a 5-minute segment every week where somebody does a show and tell on something they've done with AI.
Dedicated Slack Channel: We have a Gen AI channel where all news, thoughts, new technology, and uses go. It's self-selecting, but most of the business is in it.
AI Evangelists: We have an AI evangelist in every team - the person who self-selected, naturally reads everything, naturally experiments. They're like the go-to person for new discoveries.
Budgeting for AI Tools
Bethany: We're doing experimentation budgets, and for all of our systems of record or standard tools, as they come up for renewal, we are investigating the market and looking for AI-first alternatives. Our general assumption is that businesses with AI stuck on the side are not going to be as good as AI-first businesses.
For existing systems of record, when they're coming up for renewal, we're not doing 3-year renewals for anything. We're doing one-year renewals so that we have choice - our guess is something will be fit for purpose in a year.
Charlie Cowan's AI Journey: Building Kowalah from Scratch
Charlie: In August 2024, I was thinking, "I'm going to set up an AI boutique consultancy," and then suddenly thought, "Who am I to tell people how they should be implementing these technologies if I've not done it myself? That's just snake oil."
The Non-Technical Founder's AI Toolkit
I was WhatsApping friends in September 2024, asking if they knew any developers who could help me build something because I'm a non-technical founder. But over the summer, amazing new tools had been coming out:
Claude: New versions of one of my favorite LLMs
Cursor: An AI development environment
V0: A UI development tool
I thought, "I'll just see how far I can get as a non-developer."
Building with Claude.ai as Technical Co-Founder
I set up a Claude project called Kowalah and gave it custom instructions: "You are my technical co-founder. I haven't got a clue what I'm doing. I've got an idea for an app, but I want you to help me architect it, structure the project, understand what tech stack I should use. I'm going to come back to you with questions and you should push back on me if what I'm asking for is a stupid idea. Our goal is to get to an MVP where we can get some paying customers."
One of the great things about having Claude as a technical co-founder is you can ask stupid questions, you can ask the same question you asked yesterday, and it doesn't complain.
The Development Process
Charlie: Next step was going to V0 - an AI-powered generative UI tool. I uploaded the context from Claude (the whole PRD - product requirements document) and started designing: "Can you design the home screen? I want these sections. Here are my brand colors. I want a dark theme."
Without me having to write a very long prompt, because it already had the entire PRD, out came the sidebar, the project upload document, the chat screen. Even better, you can put screenshots of other apps you like and say, "I want that, but in my project context."
The Cost Breakdown
Brandon: How much did you spend on IT to build Kowalah?
Charlie:
Claude: $20/month
Chat GPT: $20/month
Replit: $120/year
V0: $20/month
Cursor: $30-40/month
The Emotional Journey
Charlie: I would say there were 3 weekends when I literally broke down in tears in front of my wife saying, "Who am I kidding? I don't know what I'm talking about. I should stick to doing what I'm good at."
This is one of the problems with AI development tools - yes, it gives you acceleration from 0 to amateur, but when something goes wrong, you don't know what's gone wrong because you don't actually understand what you're looking at.
I'd put on a Lenny Rachitsky podcast, come back inspired, and try one more thing - and that one thing always solved it.
From Zero to Production in 12 Weeks
It was 12 weeks from having never written a line of code to having a production app, done during weekends and a few evenings.
Two podcast episodes really stuck with me during this process:
The Google Docs founder: "We didn't know what we were doing, but throughout my entire career I've just got to the edge of what I know and fucked around. By getting to the edge of what you know and fucking around, you end up learning stuff."
Nikita Bear: "The moment you hit virality, you have to build the whole thing again because it isn't built for the scale that you need." I took the positive - I don't need to build an enterprise solution, just something that gets to the first 5 or 10 users and solves their problem.
Data Privacy and AI
The Hidden Risk of Personal AI Accounts
Charlie: One thing I'm always having to remind myself - the majority of people are not in the AI bubble. Most companies either have no AI policy or a pessimistic, restrictive one: "You must not use Chat GPT on your company laptop."
But here's the critical issue: if you're on a free or paid personal account for Chat GPT, by default, OpenAI can use what you upload (your content and files) to train the model. You can turn this off in settings, but you have to opt out, and most people don't know this.
The Solution: Enterprise Accounts
On Teams or Enterprise accounts (same with Claude.ai), by default your content is NOT used to train the model unless you explicitly opt in.
Companies that are most risk-averse, saying "you must not use AI and we're not paying for an account," actually push everything underground to personal accounts which by default get all your company data sucked up into these models.
This is no time to be an ostrich with your head in the sand. People are using this - you can choose to ignore it and create a data risk, or embrace it where the default is your data won't be used to train models.
Transforming Your Business: The 10X Employee Vision
Two Approaches to AI Implementation
Charlie: If you imagine a 5,000-person company, you could say "if we use AI we could be more efficient and have 2,000 people instead" - that's defensive, cost-saving.
The other approach: "We're 5,000 people, but if we used AI we can immediately act like a 50,000-person company by giving everyone we've got way more capability."
Benioff was talking about not hiring more software engineers this year because they can get existing engineers to be way more productive.
The Three Buckets of Work
I split work into three buckets:
Work you receive: Requirements, RFPs, contracts, regulations - things someone gives you that you need to understand, analyze, summarize
Work you do: Research accounts, prepare proposals, manage direct reports
Work you pass on: Create proposals, requirements, summaries, briefings
How can you use AI to be much more capable at ingesting information, more effective at doing your job, and higher quality/speed in your output?
Practical Example: Managing Direct Reports
Create a Claude project for each direct report. Upload:
Regular 1-to-1 notes
Performance reviews
OKRs
DISC assessments
Career aspirations
Then ask: "I'm preparing for my 1-to-1 with Beth. Can you help me think through stretch goals that support her career aspirations?" or "I need to provide challenging feedback. Can you help role-play how to deliver this understanding her DISC profile?"
Suddenly you have 10 individual leadership coaches for each direct report.
Advanced AI Tools and Techniques
Charlie: I ended up buying Claude because of its ability to mimic your writing. Claude creates my writing better than Chat GPT out of the box.
I use it when I have to be creative and I'm not being creative. For presentations, I'll say "I want an image that evokes a certain feeling" or "Give me ideas for images for 'Exceeding our target this quarter.'"
Notebook LM: The Document Analysis Game-Changer
Charlie: Notebook LM from Google takes a very intuitive approach. You upload a source (document, YouTube video, website) and it ingests that source and only that to drive AI outputs.
It creates:
A briefing document
FAQs
A podcast interview between two AI-generated hosts
Example: The EU AI Act - a 144-page dry document. Upload it and get a 31-minute podcast discussing it. But even better, you can join the podcast mid-stream and ask questions: "Sorry, I don't understand that" or "What's the first deadline I have to consider?"
Brandon: We're experimenting with taking all our sales enablement content and turning it into an internal podcast for the sales team to consume while on the road.
The Multi-LLM Strategy
Why You Need Multiple AI Tools
This is so different from traditional SaaS. Five years ago, buying a CRM or ERP, you'd pick one. That's not how it works with LLMs.
People aren't picking one LLM. You'll have a composable architecture - multiple tools, some closed source, some open source. You'll use different ones within the same workflow: this one for research, this one for writing.
From a COO/CIO/CTO perspective, think about Lego blocks. You're building a box of Lego that people can build what they want at the time. This isn't traditional procurement.
If listeners can only take away one thing: plant the seed of AI use in every single one of your business teams. I say business to mean this is NOT an IT project.
For every team, give them just one way of using publicly available tools (Chat GPT, Claude, Gemini) to improve one aspect of their work. It's those people closest to the business problem and process who will find new ideas.