A lot of SaaS teams are adding AI to their stack and hoping it shows up in output. This episode is about what it actually looks like when you build the whole workflow around it and why the thinking behind it matters as much as the tooling.

Guido Schmitz is the CTO and co-founder of OneTeam, a mobile platform that helps hospitality and retail companies connect, engage, and develop their frontline workforce. They've been building for 11 years and are shipping faster than ever without growing headcount to do it. In this conversation, we get into how they use AI across every department, how they think about product quality and delight in B2B software, and what actually happened when they stopped selling features and started selling outcomes.

🧠 What you'll learn in this episode:

0:00 - What Love at First Try is about and who it's for

0:55 - How Guido started OneTeam at 19 and what problem they were solving

5:24 - Who OneTeam is built for and why the ICP is so specific

10:30 - What the employee experience looks like before OneTeam, and why onboarding matters more than most companies think

17:10 - The €3,000 cost of losing an employee and why reducing turnover by 1–2% pays for the tool many times over

19:23 - The aha moment for new employees: introducing yourself before your first shift and getting welcomed by colleagues

21:57 - How OneTeam thinks about empty states, animations, and the small details that make B2B software feel less like a chore

25:30 - The growth challenge that comes with building a multi-product platform and how messaging for a broad product is genuinely hard

26:11 - How going from feature selling to outcome selling and narrowing the ICP doubled their win rate

34:32 - The SaaS apocalypse debate and why vibe coding your way out of a $100/month subscription usually doesn't make financial sense

40:55 - How OneTeam adopted Cursor two years ago and why they're seeing roughly 7x the code volume per developer

42:16 - How Guido uses Lovable to prototype feature improvements and get real customer feedback in two to three days

44:26 - The agentic workforce running in Slack: a marketing agent, product agent, engineering agent, analyst, and sales ops agent — and what each one actually does

46:23 - The error-monitoring agent that researches bugs, proposes fixes, runs tests, and opens pull requests for developer review

47:22 - The competitor intelligence report that lands every Monday morning — and why it would have taken five or more people before

1:01:34 - The mental model behind it all: ask what you'd do with 10x the team, then figure out if you can do it agentically

1:05:34 - How AI is changing the design process, and why the idea is now the constraint — not the execution

💡 Actionable takeaways from Guido

Steal these quick wins:

1. Start your agentic setup with one scoped job, not a big rollout

A good entry point is a single, specific job like a competitor intelligence report that runs every Monday morning. Define what the agent should look at, what it should extract, and where it should drop the output. Build confidence there before you add more.

2. Give new employees a social introduction before their first shift

OneTeam built a feature where new hires introduce themselves on the company's social timeline before they even start, and colleagues respond with a high five or a welcome message. It's one of the strongest early signals of belonging you can create, and it costs almost nothing.

3. Narrow your ICP before you touch your messaging

If your win rates feel stuck, the problem might not be how you're describing your product. It might be who you're describing it to. OneTeam went from "any company with non-desk employees" to a very specific type of hospitality brand. The messaging got sharper because the audience got sharper.

4. Flip the question when thinking about AI and headcount

Instead of asking "what can AI help us with?", ask "what would we do if we had 10x the team?" Then figure out which of those things can be done agentically. OneTeam is processing thousands of prospects a week with an agent that would have needed five or more SDRs before. The constraint isn't capability. It's imagination.

00:00 — Introduction

Jim: Hey, I'm Jim, and this is the Love at First Try podcast, a podcast for SaaS CEOs and developers that truly want to learn more about design and care about it, but have no designers and find it too complex. In every episode, we discuss how to design products that become sticky and unforgettable. We dive into the topics of taste, UX, growth, and conversions, and we share practical tips and frameworks you can add into your development process. Enough with the intro — let's dive into today's episode.

00:28 — Guest intro and OneTeam origin story

Jim: Officially welcome to the podcast. Thank you for making the time today. We always start by having an intro about who you are and what is your story, so feel free to share a bit of your personal story but also the company story of OneTeam — how you started, what the idea was, and so on.

Guido: Yeah, for sure. So I'm Guido. I'm CTO and co-founder of a company called OneTeam. What we do is we help companies turn their customer-facing workforce into their biggest competitive advantage — by connecting, engaging, and developing them via one single mobile platform. That's a really big vision from the start already. We're basically taking all these different software tools and combining them into a single platform. It took a while. We're already at it for about 11 years. Personally, I've been coding for more than half of my life. Around when I was 13 or 14, I started coding. My dad gave me a book called HTML for Dummies when we were going to France on vacation. I started making game databases and stuff on his laptop because I didn't have my own.

I started building websites as a freelancer for the neighbors — my dad knew them, they had a salon, they needed a website. We made a WordPress site back then. Now I would never use WordPress again, but it was cool at the time.

The first couple of years of OneTeam were really rough. We had a big vision, but my co-founder and I were about 19 years old. I was studying computer science and that's how we met — we're from the same village. We thought: there are so many nice consumer apps — WhatsApp, Facebook, Snapchat — but if you're working at a company, the schedule was printed out hanging in the canteen. All the news was printed on bulletin boards. You were thrown into WhatsApp groups. We were like, why isn't there a good app where everything comes together for work?

So it started really simple — a Facebook-like feed where the company could post things, colleagues could post things, people could share if they couldn't work tomorrow, and everyone could get push notifications and take over shifts. That's really what it started as. We got our first customers — supermarkets back then — and now we've really grown into this big platform where we help companies connect, engage, and develop their whole frontline workforce. About 85% of the global workforce doesn't work behind a desk. They're working in hotels, restaurants, bars. That's quite a big market.

Jim: Pretty wild, right? I forget that sometimes.

Guido: Imagine you're working. You get thrown into all of these WhatsApp groups. Your onboarding is done via a manual. You're also invited to this scheduling software and also getting surveys from different software tools — all really scattered and outdated. We're basically bringing that together within a single platform. We power about 3,000 locations, mainly in the Netherlands, and also customers in Germany, Austria, even Latvia. It's been a journey. But really exciting.

05:24 — ICP and the employee engagement problem

Jim: You seem to have a very specific ICP — ideal customer profile.

Guido: Yeah. Our ICP is basically hospitality-minded brands where the employee makes a difference in the guest experience. That's really specific. In hospitality especially, also in retail, the turnover is so high. They hire tens of new people every month. How do you make sure they're onboarded correctly? How do you make sure they are really part of your brand and culture? Because that makes them stay longer. The big problem in these markets is that turnover is so high because there's no employee engagement. The communication is really bad. They're thrown into WhatsApp groups. Managers have all these notifications on their phones at night when they're not working.

The non-desk workforce — they're not behind their desk, they need a mobile app. That's where we shine. We help companies with onboarding, offboarding, training via e-learning and mobile learnings. We also have a communication hub with a social timeline where the company can easily share news.

For example, when COVID hit, they were doubting if they should implement the tool. It was about 3,000 employees. It was a big project, but they never regretted it. With OneTeam, it was so easy to reach all their employees with just a tap — everyone gets a push notification about what's happening, what the new policies are, whether they're opening their stores or not.

Previously, how that worked: they would email regional managers, regional managers would email store managers, store managers would tell the people on the floor. But most of them are part-time, so a lot of people would never get the news. Via this mobile platform, you bring everyone together. They have groups for their own location, their own department. A visual merchandising employee in Amsterdam can share what they did, and that can reach a location all the way in Limburg. Previously that would never happen.

Our ICP is quite specific — we look for organizations with at least two locations and somewhere between 120 to 50 employees. If you're a company with 50 or 60 people, WhatsApp groups are fine. But at 200, 300, 400 people, WhatsApp groups become a mess.

09:38 — What life looks like before OneTeam

Jim: What is the life of your ICP before OneTeam?

Guido: Most of the time they have an outdated intranet that only desk people can see — people on the floor will never see it. They have WhatsApp groups for every location, maybe even per department within each location. Onboarding is very outdated — they hire tens of people a month, and it's all done via paper-based checklists: what needs to happen in the first month, do this and that, and they register that on paper. We digitize this. HR saves tons of hours, and so do managers, because the onboarding experience is so streamlined.

If you start working somewhere and your onboarding is really bad, you think: what kind of company is this? With OneTeam, new employees are invited before their first working day. They already get to know colleagues before their first shift. They go through their onboarding — what's expected of them, what the rules are, what the company is about.

We recently launched something called Performance Management — digitizing performance reviews. At most of our ICPs, this process is still done via paper. They have a paper form, sit down together, go through it, sign it, scan it, and put it in the HR system. Really outdated. And because it feels clunky, a lot of companies just don't do it regularly. With OneTeam, performance reviews are scheduled automatically in advance based on days in service. The manager and employee both get push notifications. There are nice animations — "it's time for coffee" — and we try to make the experience genuinely nice, and also not too detailed. For our ICP, performance reviews look very different than at a corporate. Less complex, less linked to KPIs.

Jim: Yeah, love it.

Guido: That's why it took a long time — we're basically combining multiple software companies into one. Sometimes we struggle a bit with marketing because we're doing so much and we want to stay lean. But now with AI, we can do more with the same team.

13:18 — The value of B2B SaaS and the all-in-one approach

Jim: A quick comment on this — one reason I love B2B SaaS is that with B2C, especially social media, it's often about stealing your attention, even addiction. With B2B, you're actually making people's lives easier. That's the real promise of design: we make your life better. The more I talk with SaaS founders, the more fascinating it feels how different every industry is. We work with ZenMaid — scheduling software for cleaning businesses in the US, specialized in residential cleaning. It's a similar story: scattered tools, a mess, people wanting everything in one place. Your approach of bringing everything into one tool — I love it. Customers just want to simplify their life.

Guido: For sure. And the data point you made is really true. We're combining so many data points that we can almost predict things — ethically, of course. We can see based on engagement whether people did their onboarding on time, whether they're doing their training regularly. People that are not engaged are most likely to leave your company. We can give companies a nudge: hey, this person's engagement is falling — maybe talk to them. And if we're only able to reduce turnover by 1 or 2%, which we easily do, the ROI of the subscription is already at least 10x.

Jim: That's very interesting. For cleaning businesses, people think the number one problem is sales or marketing, but actually it's employee retention. They turn too fast and it costs a lot to bring new people in and train them.

Guido: It costs on average around €3,000 — or dollars — to onboard a new person. In all the lost productivity, the manager hours, the HR time, training someone and getting them up to speed. So if you make sure 200 people don't leave your company each year — 200 times 3,000. Yeah. That's a big number. And we're not that expensive compared to that.

17:40 — Product quality and the aha moment

Jim: Before we dive into AI and product development, I want to discuss making your SaaS a "love at first try." You mentioned delightful details — I really appreciate when founders sweat the details. Is there an aha moment when someone signs up to your product? Products that do the all-in-one approach can make onboarding challenging. Is there a moment where people think, "oh, now I get it"?

Guido: From an employee perspective — we have a default onboarding flow. You get to know the platform, what it does. Then you're in what we call the hub — the home feed — and it's already saying "welcome to [company], glad you've joined us, here's your onboarding." It starts in a Duolingo-style way, with content about what the company is about, the history, some small games to help you remember the rules and what's expected. Then confetti, points — we gamify it a little. But that's, I think, still functional. I'm not sure if the employee gets an aha moment from that part.

I think the biggest aha moment is probably our feature where every new person at the company introduces themselves to their location and gets "high fives." We don't have likes like Facebook — we have a high five, because it's a team feeling. We've also added other reaction types because at some point, tragic news was being shared on the social feed, and a high five in that context felt a bit weird.

Before your first working day, you introduce yourself, it's publicly shared on the social timeline, and colleagues start commenting: "Hey, welcome — glad to see you, see you soon." And you haven't even started working yet. That's already the first start of bonding with your employer and your team. I think that's probably the aha moment.

But then we also spend a lot of time on edge cases — animations, empty states. When we started 11 years ago, B2B enterprise software was really functional. We still see that — we integrate with a lot of HR and workforce management vendors and it's really functional. There's not much product taste added. We try to ask: how can we make it not just functional? Instead of "no courses found," we make a nice animation: "It seems you're all up to date!" with confetti. That's a whole different way of looking at the product. In B2B software that's changing a lot now, but still a lot of work to do.

25:30 — Messaging challenges with a multi-product platform

Jim: Last question before we dive into AI: what's been one of the biggest recent challenges in terms of growth?

Guido: The main thing we've been doing in the last few months to years is pivoting on our go-to-market story. We went from feature selling to really outcome selling, and we narrowed down our ICP significantly. Before, we would say we're for every company with non-desk employees — including retail companies with just one location. That made it really hard to let the customer know who we're actually for. If you're for everyone, who are you actually for?

Now we've narrowed to really the customer-facing teams where guest experience makes a difference for the company. That's the whole strategic angle. And it seems to be paying off — we directly saw it in our win rates, which were doubling.

But it takes a lot of time. You don't know what you don't know. You have to be really ego-less in terms of trying to change direction, because the market is letting you know if it's the right direction. If it's not, you won't grow easily.

The fact that we are such a big platform — we bundle maybe three or four software companies into one app — makes messaging hard. If we were just an internal communication tool, marketing would be so much easier. But we have onboarding, offboarding, surveys, e-learning, performance management. That's, I think, the biggest struggle we had. But it seems to be paying off now. And it's a constant process — you just keep improving the messaging every month until it's good. If it's ever good.

Jim: The whole idea of niching down is something I'm a big fan of because it's one of those things where business strategy has an impact on literally everything. You could say your messaging isn't working, your marketer isn't good — but often it's something deeper. If it's not clear who you're serving, you can't generate clear copy. Once you improve specificity, everything starts getting more focused: product strategy, growth strategy, feature decisions.

Guido: Especially now with AI, a tool is just a tool. Everybody will easily make tools. Maybe in two years, every point solution is a big platform. So how do you differentiate if everything is a tool? I think now it's really about white-glove support that we're implementing, the brand, and how focused you are on the outcomes of the customer — instead of just saying, hey, you can sign up here. Our competitors can all use AI as well. They can build everything we have. So what's the difference? That's also a really interesting thought experiment to see how your company has to evolve.

32:03 — The SaaS apocalypse debate

Jim: Good time to jump into AI. But let me start with AI inside your product. There's all this paranoia — the SaaS apocalypse, vibe coding your tools instead of paying for them. My personal take is that it depends entirely on the scope of the tool. It's not easy to build OneTeam — that's 11 years of iteration and deep thinking. Getting to a V1 is getting easier. But maintenance, bug fixes, iteration, getting feedback, who owns the internal tool when colleagues start making requests — suddenly you're a CEO of an internal tool. Maybe in large enterprises you'd see a small internal team build their own SaaS, but even there — compliance, data protection, security — these aren't easy to replicate with two or three people. In the end, businesses still need tools. These tools are just getting smarter.

Guido: I saw all the engagement-farming posts on this too. I think the right question to ask is: who do you trust more to give the outcome we are delivering? We help companies make their customer-facing workforce as effective as possible. We learn from all our customers, we see what's happening in the market. Based on all that, we build the product. If you vibe code your own version of OneTeam, what do you build next — based on what knowledge?

We're also quite mission-critical. Operational internal communication is happening within OneTeam. Do you want to build that yourself, maintain it, take the risk of it breaking? Would you vibe code Slack internally? If it breaks, you'd be in a hard time.

That said, I think small point solutions might have a harder time — but those are mostly not that expensive. Let's say you pay $100 a month for Calendly. You could probably vibe code it. You can get to 80% in a day or two. But that last 20% might take four or five times as long. By the time you calculate that against your hourly wage, you've maybe pre-paid four years of the Calendly subscription. And that's not counting the opportunity cost.

Jim: That's a beautiful example. Is it worth it? You can spend that week generating five times more revenue instead.

Guido: Time on business is way higher leverage anyway.

40:12 — How OneTeam uses AI internally

Jim: Now let's dive into the design and development process. How are you using AI internally at this moment?

Guido: We've been using AI in our development team for about two years now. We started when Cursor came out — I immediately got everyone a subscription to get into the mindset of the new paradigm. It started as a pair programmer, like a rubber duck you could talk to when things were breaking. But the models back then weren't that good. Now our AI workflow is adopted across the whole company. Our tech leads are writing all the user stories with AI, checking them themselves, and putting them into Linear. People are using multiple terminals in Cursor, plan mode, everything.

Compared to 2023, we've seen seven times more volume per developer in terms of lines of code. Lines of code isn't the best metric, but it's better than pull requests in my opinion, because PRs can be big or small. The direction is clear — we're releasing quicker and quicker.

42:16 — Prototyping with Lovable

Guido: In my role as CTO and product owner, what I really find cool is using Lovable for prototyping. When I see a problem with customers, I screenshot our product, put it in Lovable, and Lovable recreates the whole screen. Then I just iterate — my thoughts, things I want to change. It creates a Loom video-ready V2 of a feature. That can happen now within two or three days. Previously it would take so long — you sit down, make wireframes, talk with designers, then iterate, and each iteration step was maybe four or five days. Now it's less than a day.

Why Lovable and not Claude Code? The visual thing. I can just copy a screenshot. I don't need to spin anything up. In the prototyping phase, I also don't care about our design system. Obviously it will eventually be translated into the design system, but to validate a problem and a solution, the design system doesn't matter. The customers don't care either.

Jim: It's not about the visual design — it's about the UX and the specs.

44:26 — The agentic workforce in Slack

Guido: What we're doing recently is we have our own agents running in Slack. We have a full agentic workforce, built on top of OpenClaw. We have a marketing agent, product agent, engineering agent, analyst, sales ops. They're running in real time, doing customer research, transcribing all our internal meetings.

Each agent gathers its own insights, makes notes, and shares them with each other. Product changelogs are being written with these agents. When we release something, we have an agent checking: did we really release something? It cross-checks all the meetings customer success had with customers where a specific feature was mentioned. The agent knows: we released this three months ago, you had a meeting with this company, you should inform them about the new feature.

It's like having an AI colleague that augments each department and helps with the gaps. If you're growing and have more people, things slip through — but these agents always see it and give notifications.

46:23 — The error-monitoring agent

Guido: We also have agents checking our error logs. If there's an error, it researches what happened, whether it's related to a release. If not, it proposes a fix. If a fix is found and a test passes, it creates a pull request — which then has to be checked by a developer. As a technical person, I'm just like, what is going on? Things are happening automatically, agents are talking with each other.

When we release something, it creates a product changelog, which helps the marketing agent draft social media posts and creates a content calendar to help colleagues post on LinkedIn — fully reducing writer's block by giving them all these insights.

47:51 — Competitor intelligence and the "10x team" mental model

Guido: A really practical example: we have an agent making a whole intelligence report on competitors and market trends every Monday morning. Previously you would maybe have an intern, or five people scrolling through websites looking for news — customers expanding to other countries, things like that, which is valuable intelligence for the sales team. We automated all of that. We work with a lot of VAs and that work is now done by agents.

On a personal level, I'm really into this movement where I have personal assistants that read my email. I integrate my Garmin data and the agent interviews me every Friday while I'm walking — it looks up my calendar and task management, interviews me about my week, and helps me draft content for the next week.

We also automate a lot of sales operations. Every company gets researched, our CRM is always up to date with the people working at companies we want to target. We use tools like Clay. With our ICP now so narrowed, you can run research agents and do advanced prospecting with AI in a way that wasn't possible before.

What I find interesting is I still speak to recruiters and developers who are debating whether they need to use ChatGPT or Copilot. You see some companies going really fast and some still quite behind.

52:15 — Why OpenClaw

Jim: Why OpenClaw specifically? You mentioned you started on Claude Code — what's special about OpenClaw, and do you mainly interact with it through Slack?

Guido: We started building agents on Claude Code. You can make all of these skills — each agent has skills, you can set up cron jobs. Every Monday morning the agent runs, analyzes last week's meetings, or finds news websites important for your company and extracts the data. But I was reinventing the wheel in terms of permissions and security — which people have access to which agents. We also have a management agent writing investor updates and executive reports based on company-wide meetings. OpenClaw already has all of that in place: Slack channels, security protocols, a control UI where you can easily debug what went wrong. People in our workspace can just tag our agents and talk to them.

It has access to a lot of tooling: our product database (read-only, only specific tables), Apollo, Intercom, GitHub, Figma MCP, CRM, Datadog, Linear.

What's powerful with OpenClaw is the memory management. Every day the agents process what happened, make a summary of conversations. I'd rather add things on top of that framework than build it ourselves.

The other important thing is that it's just a framework — I can plug in any AI model. Our product agent might use Opus, our marketing agent might use GPT. In six or twelve months there might be an open-source version of the current level of Opus 4.6, maybe ten times cheaper. We can then easily swap models based on the tasks, and we're not vendor-locked with Anthropic.

It also learns: I asked the agent to introduce itself, and asked all my colleagues to introduce themselves too. Now the agent knows if our tech lead is talking to it, and tailors answers to his role.

1:01:34 — The "10x team" mental model

Guido: AI is helping you think bigger. Ask yourself: what would we do if we had 10 times the workforce? And then figure out if we can do it agentically. That's a really interesting thought experiment — because it's actually possible. We're going through thousands of prospects every week with an agent. How many SDRs would we have needed before? Maybe five or more. It's also the boring work — go on LinkedIn, find the latest news, find the current contact person, find their email, write a personalized message. Now the sales team can spend time actually calling people, meeting them in person, going to events. That's the high-value work.

Jim: Yeah — sales is not about spending 10 hours a day on LinkedIn fetching data and putting it in spreadsheets. That's brainless work. The thing that stood out to me about OpenClaw is the framework thinking — that with AI, the value compounds. You build, build, build on top of something. In the past, knowledge was built on people — they left, the knowledge was gone. Now you have a system that's always in the company and the whole team can tap into it. And you're not locked to one LLM — different models for different use cases.

1:05:34 — AI in the design process

Jim: On the design side — you mentioned coding and sales and marketing. Is there a specific way you use AI for design?

Guido: The main way is Lovable, like I said. It comes up with some interesting flows, but that's also because of how I prompt it — I know what good flows look like. AI fills a lot of gaps, especially edge cases, because I tend to think about the happy path only. Then we pass it down to our design team, who puts it into our design system.

Our designers are now more focused on converting concepts into the design system — the UI side — rather than doing product thinking. The main concepts come from us because we talk directly with customers. Our designers don't have customer interviews. That's done within our tech leads, customer success, me, and my co-founder. We don't have that many layers, so we can move fast.

We're also questioning what Figma's place is going forward. We already have Storybook with all our components. Technically we could go directly into the app. We still keep Figma for some nice visuals, eventually for marketing, but prototyping I do in Lovable.

Jim: On our workflow — we build in the real codebase, not in Figma, not in a separate repo. The realization we had is that prototyping on a real branch is not the same as shipping to production — that's much more complex. But building in the real environment means you see things you'd never see in a fake prototype. For example: building the customer portal, I realized the admin settings were scattered across eight different settings pages. I wouldn't have seen that on a fake prototype. We use Figma only for hardcore visual UI design — exploring a very specific component visually, playing with pixels. When you want to build a product, not play with pixels, Claude Code is the way to do it. My workflow is about 80% AI, 20% Figma.

Generating code doesn't mean you can ship that code. We see the prototype as a prototype — it just happens to be on a branch instead of a page in Figma.

Guido: Yeah, for sure. I'll probably start experimenting with Claude Code directly in our repo. The good thing is we already have our Storybook and design system quite structured — otherwise Claude would probably create a lot of new components and blow up the codebase. The thing I was afraid of before was that our codebase is so big — but the models now really figure out how your codebase works. Previously with Claude Code you had all these CLAUDE.md skill files where you had to describe your framework, your structure, how you work with databases. Now you don't need all that. It figures it out.

We're also lucky that most models are trained heavily on TypeScript and JavaScript — and we're fully TypeScript, back-end and front-end. I'd also recommend everyone to use Context7 for getting the latest docs via MCPs.

1:22:02 — Closing

Jim: Awesome. Thanks a lot for your time, Guido. It was a beautiful episode. Many cool insights.