STOP Waiting for AI. Start EXPERIMENTING ft Steven Kenway

[0:03] All right, looks like we’re live here. Thanks everyone for joining. Welcome back to Out of the Weeds Live. My name is Nish Sampath. I’m the founder of Switch Advisory.

[0:15] Today we’re talking about AI in your small and mid-size business, because the business owners we’re talking to these days are concerned about keeping up with the trends. They’re reading about AI but outside of that, almost none of them know what to do with it come Monday morning. There’s a flood of information. The speed of change is crazy and it’s causing this kind of perfect storm that’s leading to analysis paralysis.

[0:41] You’ve got hundreds of vendors pitching you and it’s frankly overwhelming, and so you sit on these decisions thinking that you’ll pick it up in the next quarter — but then a year goes by and you’re still stuck at go. So today we’re trying to get past the hype and talk practical use. Where do you actually start? What are some of the simple workflows you can look at? How do you get your team on board?

[1:00] I’ve got my friend Steve Kenway with me today. He’s a fractional CTO and the CEO of Rolling Wave Consulting. Steve has built and sold his own software company, he advises boards on tech strategy, and frankly he’s the guy owners call when a project is off the rails. Steve, thanks so much for joining us. Tell us a bit about the company.

[1:24] Yeah, thanks for having me, Nish. So Rolling Wave Consulting is about 15 years old and it was born out of the identification that to get tech projects done, leading with project management increases the likelihood of delivery success. And so although we tap into technical advisory and technical roles, we ground everything in project management. Whether we’re working inside of a Fortune 500 company or with a startup, we always go back to those principles. From a service perspective, we have purely just project managers available to help lead some of those initiatives, but we can also be your full team. And a little bit of what I’m hoping we talk about today is what does that full team kind of look like in this new advent of AI?

[2:25] Yeah, absolutely. And it really fits in frankly with our fractional models — fractional COOs plus fractional CTOs on your side of the fence, along with people to help execute that. That full delivery model makes a ton of sense in today’s evolving world and evolving technology landscape. So let’s get into it.

‍

Chapter 1: What’s Causing the Analysis Paralysis

‍

[2:48] Let’s talk about the analysis paralysis. You and I have been chatting about this for a while. What are some of the major things you’re seeing that are causing these owners to get stuck right at the starting line?

[3:01] Yeah. I think the big thing is we’re three years into this journey of generative AI and agentic AI and the power of LLMs. I think it’s the status — about 50% of the population is using the tools now. And I believe that’s the fastest to 50% in history in terms of new tooling, new technology. So everyone sees it, they help it write an email, they ask it a question, it responds in a very human-like way. But then it’s taking that and turning it into productive use and then getting that to proliferate throughout your organization.

[3:41] And the difficulty that we have as businesses is — in the early stage if you’re a startup, you’re kind of in that exploratory mode, but when you’re a business you’re in the exploitation stage. You know what your product is, you’ve got your people — it’s about how do we tune things as tightly as possible so that we get the most out of it, we get the highest quality, the best returns, the most predictable results? And so now when you bring in something like AI where a lot of owners can see the power of it, they hit that analysis paralysis getting into let’s roll it out.

[4:15] And I think there are a couple of things that are a bit different than previous technological advances. So if we look at the SaaS period — before it was very, very opinionated. It was like you come on, you sign up for Salesforce, you go through an implementation, there’s a lot of flexibility, but at the end of the day here’s how you use Salesforce. You’ve got to stay within its guardrails and they’ve got a really good playbook on how you can get up and running and maximize that. So I think that’s the biggest kind of analysis problem right now — it’s such a blue ocean. There are so many possibilities. Where do you start?

[5:01] Yeah, that’s great. Like you had a product back in the day — a SaaS product — and the user experience is very specifically outlined for you already. Here’s exactly what the workflow is, here’s how we’re going to implement it for you. Yes, there’s some minor variability, but it is what it is. Problem, solution. Now we have solution — all sorts of problems. Which one do we even tackle?

[5:36] Yeah, exactly. It reminds me — you know when we were colleagues at a large healthcare provider — we saw the back side of this, but it’s the advent of the Access database where all of a sudden you could create a database for all of your data management needs. When we came into the organization, there were hundreds of Access databases throughout it and everyone had spun those up. They were identifying a need and a problem and they were matching it and creating those databases. The problem that created on the back end was now you had this unmanaged process. Unmanaged data everywhere. It wasn’t secure, it wasn’t centralized, there were no backups because it was all just local copies. So I feel like we’re in a similar type of period where we’ve got the power of a database now in every user’s hands and we need to figure out: okay, how are we going to roll this out so we don’t end up with hundreds of Access databases again?

[6:44] Well, that’s exactly it. Like from an organizational perspective, no one’s not using it — they’re just not using it collectively and consistently. It’s kind of everyone’s personal research assistant and writing assistant and everybody’s doing whatever they want with it, putting whatever corporate information in it, and the risks are high. And that’s probably what’s leading to a lot of this fear from owners: hold on, what are we doing with this thing? It’s great that you’re experimenting independently with it, but the potential risk is also significant.

[7:26] Yeah. And we’re seeing a lot of organizations making decisions on rolling it out based on that as their guideline. So not maximizing the power of the capabilities — staying on the leading edge with OpenAI and Anthropic. They’re rolling it out with Copilot because they’re already a Microsoft shop and they feel really confident in the security mechanisms they’ve put in place. Copilot can go into that trusted environment and connect with that data. But the flip side is — right now, maybe not in the future, it’s moving very quickly — Copilot is probably the least powerful to be that transformative agent inside of your organization. It’s good for writing an email, answering a question, maybe helping you stitch together some data. But it’s nothing compared to that true agentic nature that you would get with using Claude Code, Claude Co-work, or OpenAI Codex and ChatGPT.

[8:26] Yeah, that’s actually a good point. And when I think about a lot of high-risk businesses, they are specifically leveraging the Microsoft platform just because of their fear from a security perspective, even though the tools aren’t necessarily the bleeding edge. I was speaking to some lawyers yesterday and that’s exactly where they’re going. There’s just too much at risk for them at this given moment given what they know.

[8:56] Yeah. And it’s something you have to understand — your risk threshold. And I think if you come up with a strategy that takes that into consideration, but you’re acutely aware of how you want to get to that next level — and the next level is really truly getting into that agentic state where you’re not just prompting this thing and it’s giving you responses back — research, writing, etc. — but it’s actually going out and doing work for you.

[9:30] Which is really where we all want to be. That’s exactly it. Today we’re still for the most part stuck in the research and writing mode. What does an actual agentic workflow look like inside of a business? Give us some of these use cases you’ve seen play out.

‍

Chapter 2: What Agentic AI Actually Looks Like

‍

[9:47] Yeah. So an agentic workflow is essentially this recursive, autonomous set of responses and queries that are happening inside of the LLM. And what has enabled that — the LLM remains the same, the brain is the same, the model is the same — it’s the harnessing around it that’s allowed you to stitch together and create reasoning. So you can give it a task and it can go away and work through the steps — three, four, five hundred steps in that process — to truly get some work done.

[10:21] So where are we seeing some really cool value? There are kind of two heuristics that I use today. Number one is: is there a task that you’re doing that’s very repetitive and you haven’t necessarily had the technical expertise to automate? One of the things that LLMs are really good at — especially coding agents — is writing code. And so now, like never before, an everyday person who doesn’t write code can prompt Claude Code to go in and write code and take this repetitive task that you do all the time and turn it into automation.

[11:08] So that’s the first one. The second one is: it’s the thing I never have time to do but I want to do because it would make my business better, but I just never get around to doing it. We’re busy, things come up, fires. So you just don’t get to it — which is most owners and most of their time. You’re dealing with the daily grind, you can’t figure out how to work on the business and get out of the weeds, and you get insular and focused down and you don’t get the time to work on the cool things.

[11:40] Exactly. And so one of the big ones that I’ve seen — a real quick win, extreme value — is being able to use a coding agent to build out your reporting. The old process was: you go in, you download all these spreadsheets, you export all the data from different places, you’ve got a spreadsheet that you’ve built, and then you go in and plug in the values or copy and paste from the original source, and then ultimately it gives you that view of your business — that dashboard view.

[12:17] You know, you work in a big Fortune 500 company and you’ve paid people to put all this data into an enterprise data warehouse with PowerBI plugged in and they’ve already designed all the dashboards. But if you’re a small-medium business, normally you’re doing that yourself at the end of every month with Excel. And I know in my case — I try to stay as disciplined as possible — but that’s one of the first things that gets cut. I’ll do it next month. And so you fly blind for a month.

[12:51] And we’ve started using in this particular case Codex to go and assemble all that reporting. And one of the really cool parts about it is that because coding agents and co-working tools are really good at writing code — whether it’s single-use or something used consistently going forward — it can connect to APIs really easily. And so now you’re not just going and downloading the exported report that’s available in that tool today. You can hit the APIs and get the raw data.

[13:26] So in one project I’m working on, we get all of the subscription data, all the transaction data, all the patient data, the payment data — suck it all out — and then we assemble that with usage data, combine the two sources, and then it produces a monthly report that has way more depth and breadth than what we were able to do before manually. And then I also tell the LLM: go and interpret that data for me. Tell me what you think. What am I missing? What blind spots do I have? Are there any trends that you see that I don’t see? And so you kind of have this little financial analyst sitting at the end of your desk doing this work for you and giving you opinions. And sometimes the insights are really interesting — things that you didn’t think about. So that process of stitching together that reporting is a really, really easy place to start.

[14:26] Yeah. I mean that’s a huge one — internal and external. A lot of our professional service clients — that’s one of the use cases we work on first. Aggregating all that information is incredibly time-consuming. It’s taking people away from their current jobs. The real conversation is about what are we doing with the data and how are we using it effectively. I was chatting with some folks in wealth management and some clients in marketing agencies the other day, and they have to report to their clients on a routine basis. It’s a massive amount of time-consuming work. Just being able to pull that data together, provide that overview so that they can then provide the human interpretation afterwards — adding that personal touch of really knowing their client — that’s a massive value add.

[15:29] Exactly. And you could do that for multiple stakeholders with a separate report for each of them based on what’s most important to them — all on the same foundational information.

[15:42] The other cool thing is — one of the downsides of LLMs is that they’re non-deterministic. So you run the same prompt five times, you’re probably going to get five slightly different answers, just the way the math works. But you can start to take some of the prompts and turn them into skills and runbooks accompanied with code, so it becomes more deterministic. So in that scenario you mentioned — if you’ve got to run these reports each week or each month and they all care about slightly different things — you can create a runbook so it’s not you having to go through that same agentic prompting process each time. You can just say, “Hey, run this skill, run this runbook, and produce this report for that particular stakeholder.” And boom, it’s done.

[16:32] I like that a lot. And I think a friend of ours posted this morning on LinkedIn talking about how AI has created so much efficiency but somehow none of us have got our time back from it — we just found ways to fill that time with more tasks and more AI work. But this feels like one of those use cases that is actually effective at removing some of the human intervention required and cutting many, many hours of unnecessary overhead time.

[17:06] Yeah, absolutely. And I think the other thing too is — it all depends on your harnessing. If you feel like you’re not gaining time from this, one is you’re probably doing things you just wanted to do but couldn’t do before. But you also have to get your environment set up. So two elements there.

‍

Chapter 3: Setting Up Your Agentic Environment

‍

[17:24] One is you’ve got to truly be on an agentic harness — meaning you’ve got to be running a framework underneath that allows you to truly interact with your agent like they’re a first-class citizen. Like they’re an employee. So personally, I have OpenClaude running on a Mac Mini in a secure sandbox so that the agent employee doesn’t go rogue, but I’m able to send it text messages and tell it to do work. So if I get a research idea, I can assign it, it can work and churn and do this work overnight while I’m sleeping, and I can come back and see the report in the morning.

[18:20] And so I think a lot of folks who aren’t finding some of those advantages are sitting at the terminal at their computer and they’re the ones actively prompting face to face with their agents or their LLMs. The second piece is starting to pull those agents into your team. There’s a new platform that came out about a week and a half ago called Buzz from Jack Dorsey at Block. And one of my university cohort colleagues works at Shopify now — he was also part of the startup that I exited — and he was telling me that within Shopify, they’ve built their own platform. They bake all their own tools.

[19:02] And so the whole concept behind this is that your agents are first-class citizens. It’s very similar to Slack where we can all chat with each other, but right there in the conversation is an agent — and they are as a person. And so what this allows us to do is bring our agents into the teams. So imagine you had a group chat where you’ve got two humans and two agents and maybe they do an operational daily check — they run through customer care as an example and just see: are there any important messages, anything that we need to respond to? You can assign that work to the agent to do all the prep work.

[19:47] And because the humans are also in that, my colleague — let’s say you — I prompt Rick, my agent, to do the work. You can see the result. Nish can see the result of the thinking of the agent working through it. And let’s say the other agent is Bill. Bill’s going to draft a whole bunch of responses based on Rick’s work. So Rick can tell Bill to do the work afterwards. And so now you’ve got that agent-to-agent communication. Meanwhile, Nish and Steve are still in there just like team members. We can monitor their work, see the activities occurring, intervene if necessary.

[20:26] But a lot of it today is — as I said — terminal to terminal. People have their personal agent. The only time you ever get the output of that is if a human then turns around and gives it to the rest of the organization. So how you set up your harnessing and your environment is going to be critical going forward so that you can really pull your agents in as first-class citizens.

[20:52] So think about it almost as a team, right? In that analogy, I’m almost thinking like morning scrum comes around — you have your agents plus your other team members on scrum. Everybody’s having a conversation, digesting some of that information, talking tasks and priorities back and forth. Is that a world we’re looking towards?

[21:11] That’s a world we’re looking towards, and it’s going to go a level above that. So in Canada we have something called Scientific Research and Experimental Development — or SR&ED — and you have to keep extremely detailed records to get those tax credits from the Canadian government. You could have in your daily scrum a SR&ED agent. As people are talking, it’s going back inside of Jira and validating: is all the time tracking there, are all the notes there, is everything connected to the code? So that at the end of the year when you get to your SR&ED submission, you’re doing your tax return, you’ve got all the underlying evidence to support receiving that tax credit with no additional effort, no project work. Maybe you still bring in the consultants to help you characterize it and protect your interests, but all the pre-work that’s normally so time-consuming is all done for you in the moment in those scrums by your SR&ED agent.

[22:10] And so it’s taking it a leap beyond what we used to do — like grant writing or proposal writing. Cool, you can leverage your historical proposals. But now you can almost say: okay, let’s go measure the efficiency of the last project that took this amount of time. How many human hours? What kind of financials did we have? What was the impact on the business? Pull all those data sources together collectively, and then have these tools write out the information. But someone’s monitoring in the back end saying: as the work’s happening in the last project, are you producing the right data so that we can use it for the next go-around?

[23:01] 100%. Yeah, absolutely. And then it even goes a layer on top of that. So once you’ve got that good hygiene within your teams — again it’s always tough to get your teams to do that naturally, but if you can have someone behind the scenes reviewing, monitoring, maybe even doing some of it for them — you’re in a really good spot.

[23:18] But then when you go to write the actual proposal, the other big thing that I do is I hardly type anymore. Everything is voice. I think I maxed out in grade four at 40 words per minute in typing, but I can definitely talk a lot faster and I can provide a lot more context. And when I’m crafting a proposal, for example, I can really inject the nuances of what I want to say — whether it’s responding to an RFP that’s formal or something where a customer has just come to us and asked for a proposal or a solution. Using voice really helps me get that context in my voice. But the agent can then go and fact-check against all that underlying foundation information and then write it in my voice so you don’t get AI slop. Because I think we’ve all seen it now — you can see AI slop from a mile away.

[24:21] Totally. And there are techniques where you can use a lot of this tooling to get the advantages, get it to do the work, without producing AI slop.

‍

Chapter 4: Process Hygiene — The Foundation for AI to Work

‍

[24:32] Let’s talk about the hygiene thing because I think that’s very interesting. Is it the human that needs to support the hygiene or is it the AI that’s supporting the human’s hygiene? What’s that relationship?

[24:43] Yeah. So I think at the onset of all of this, you’ve got to have good business processes. You’ve got to have good workflows. So we humans design the process in such a way that it’s mapped out, everyone understands what their role is in the process. And — getting a little bit on my soapbox — people don’t make mistakes, there’s just bad process. So making sure that we’ve got a good process in place. The same applies to agents: if you’ve got a good process in place, your agents are less likely to make mistakes and produce slop.

[25:14] And so on the hygiene side of things — I’ll give you one example because we were working on it the other day. We’re working on a piece of code right now where every time we change the API, we want to make sure that the documentation — which is automated — is updated at the same time. Historically what you would have done is put in a blocker: we’re going to block this particular change until the developer has gone in and done that thing and then moved things forward. That can create a whole bunch of other problems. But what we can do now through an agent or a skill is actually tell the agent: as it’s going through the workflow in the process and when the agent does its code review and its checks, check to make sure that the documentation’s up to date. And so it can really support that hygiene side of things, as long as we’ve designed the workflow or the process in such a way that the agent understands the input, understands the logic, and has the right output to send a response to the humans or other agents to act upon it.

[26:24] Okay, that makes a ton of sense. Apply that to the bookkeeping use case you had told me about. How does hygiene fit into a scenario where you’re having the tool do reconciliation for you?

[26:37] Yeah. So my dad is a 76-year-old — he was a CA before there were CPAs — and he helps with our books. One of the things that he self-discovered on his own was using Claude Co-work to do reconciliation. So we have to bring in all our transactions. He would do this all manually with a VLOOKUP and a spreadsheet. And now he just at the end of every month — on top of all the other accounting and bookkeeping stuff — he goes and attacks the problem with Claude Code, puts all the underlying materials into a folder the same way he did it before, and then gets Claude to do the actual reconciliation.

[27:24] And so what it’s helped us is — we always did the reconciliation. It would happen I think twice a year. Previously we would 100% do it at the end of the year before we closed the books to make sure there were no outstanding accounts, no dollars being left, our payment processor wasn’t stealing any money. But now we can do it every single month. And so the health and the confidence in our accounting and our bookkeeping is so much higher. And honestly, it’s not a huge change in terms of implementation because it’s the same workflow and same process that we’ve followed all along. It’s just that we can do it more frequently.

[28:13] You can trust the numbers again.

[28:15] You can trust the numbers. It’s pretty amazing.

‍

Chapter 5: What Does It Actually Cost and How Long Does It Take?

‍

[28:20] So we’ve talked about some of the use cases. What does it actually take to stand something like this up? Realistically — who owns it? Is it a 12, 18-month initiative? Hundreds of thousands of dollars? I think those are some of the fears that people have before they even cut into it.

[28:38] Yeah. And in projects of yesteryear in the SaaS world, the supply of writing code was rate-limited. We are now in a new supply abundance in terms of writing code. One of the things that we used to talk about as software developers is you never wanted to have single-use code — that was a waste. Why expel all this effort if you’re only going to use it once? Right now with coding agents, single-use code is free and cheap, so you might as well use it.

[29:06] And so what it’s done is — because it’s just a force multiplier in being able to generate code and use LLMs — it’s taken a 12-month project and shrunk it down significantly. Or it’s a 12-month project where you’ve been able to dramatically increase the amount of scope, depth, and complexity of what you’re able to accomplish.

[29:37] And there is one caveat I’m seeing though. You have to make sure that the process is guided. Because as much as it’s a force multiplier in a positive direction, it can be a force multiplier — like the Access database scenario — in the wrong direction.

[29:57] And so one of the things — the model we’re moving towards — is having forward-deployed pods. A lot of people are probably familiar with the forward-deployed engineer term, where you take an engineer who understands this and put them on the front lines to help go and integrate the reporting for that staff manager, for example. This takes it a step further and pairs that developer with a PM.

[30:24] And the PM is responsible for liaising with the business, coming up with the backlog of things that need to be attacked. And the PM can do a little bit more BA work than in years past because they have the tools themselves to do a lot of this agentic work. So they’re truly managing the relationship, understanding those workflows, identifying the pain points, and then keeping the engineer focused in the right direction so they can go really quickly. And so those 12-month projects — whether waterfall or agile — where you’ve got multiple sprints, you can start to see value really quickly. You can see value in 30 days. So that’s really the way we’re trying to position it for organizations. It’s less about this big piece of work and it’s about making sure we stay true to the business and tackling as many of those problems as quickly as we can.

‍

Chapter 6: A Real Use Case — Manufacturing

‍

[31:32] So let’s take a manufacturing organization as an example. A manufacturing company comes to you, we look at their value chain, understand the products they’re delivering and what’s going on. Where does the team start?

[31:44] Yeah. Let’s say since I’ve worked a little bit in the service and repair side of things — what we would do is say: okay, let’s focus on that particular problem. And I know that’s traditionally been especially difficult depending on the size of the manufacturing. If you’re a multi-global organization, it gets complicated very quickly, especially if you’ve got to deploy service techs out to stuff that’s been manufactured and installed — large kilns, medical devices, etc.

[32:22] So what I would start with there is: okay, let’s go in and understand the problems. We would start with discovery with the PM. The engineer would be there to hear all the detail, get all that context, and start identifying those highest pain points. And I know that in the case of service and repair for large complex manufacturing scenarios, it’s getting the underlying data back to the service organization to then be able to assess: okay, what do we need to do here? Do we need to ship them just a replacement? Do we need to fly a tech out? Or is this a hey, maybe try restarting it and we can just give you some instructions to solve your own problem?

[33:06] Unplug, replug.

[33:08] Unplug and replug. So where we could help with that is — there are a lot of different places, different techniques to get that information. In some cases the device has it. So can we build some integrations? Those things are often available in other places, other uses, but service and repair doesn’t have access to it. So can we write some integration against that?

[33:36] The second piece is, you know, traditionally in some of those SaaS products, very rigid forms that they would have to fill out — not always the most intuitive and friendly. Could we turn that into an intelligent chatbot that is prompting the individual to provide all that information, taking them through a more intelligent and human-like process of describing the problem and sharing those details? And then having an engine once you get all that accumulated to do the triage.

[34:07] So maybe it’s looking at a history of hundreds of cases previously and understanding the successful outcome. Most organizations have that now after running SaaS platforms. Using that as a knowledge base to then inform the decision-making: this is probably a warranty replacement scenario, this is probably send a tech out, this is probably just respond with this — and starting to sort a lot of these things so you can get responses out to your customers a lot faster, especially in high-volume, multi-complex environments.

[34:38] Makes sense. And you know, even for smaller manufacturers that we’re working with — they would also have the issue of not just triaging the fix, but how does that match to the math? Is it more effective and cost-efficient to send a human versus just send a new device? What does it do to our brand? How do we collect that brand information and collate all of that together? What are the survey results from our clients saying about our service? I think the real power is how do you aggregate this information that we have all these disparate sources for, and then be able to tell a more interesting story and produce really cool results from it.

[35:24] Yeah. I didn’t even think of that — like the financial calculation side has always been done, but how does it affect brand? How does it affect your NPS score? That would be another layer. And the beauty of forward-deployed pods is they can go and do all the integration work to get all that data in the same place. And the power of the LLM is it can understand all information at once, so it’s really easy and fast for it to say: okay, I’ve got to check all six of these sources, take all of these things into consideration, and then make a recommendation or triage or direct it.

[36:05] And at the end of the day, you still want humans in the loop. You’re not using this to remove humans. You’re using this to get your humans to do higher-value work — and in most cases, touch other humans. Create more human-to-human contact so they’re not having to do all this paperwork behind the scenes. You can get them on the phone with that particular person who’s upset and have a conversation, then review the notes and make that decision on how you solve things.

[36:37] That’s a really good point. So when we’re thinking about a lot of this, my head says a lot of this dies when there’s fear, uncertainty, and doubt across the team. How do you get the team on board to actually use the tools instead of fearing things like layoffs right away?

‍

Chapter 7: Getting Your Team on Board

‍

[37:02] Yeah. I think first of all there is an organizational change management piece — and I don’t think that the large foundational model companies have done themselves a great service. They’ve done it to raise money and market like these things are unbelievably powerful and going to change the world, but that’s also created a lot of the fear that it’s just going to remove people from their jobs — which I’m not seeing at all. I’m actually seeing the reverse. It’s actually starting to create more business. The companies using it are generating more revenue, generating more opportunities, and therefore there’s actually a requirement for more people.

[37:39] But the best way is organizational change management. So as you roll these things out, follow that communication plan: awareness, desire, knowledge, activation, reinforcement. Hey, we’re starting to take advantage of this. There are going to be forward-deployed pods inside of our organization to help us do that. Hey, here’s how you use these new tools and new things that we’re building for you. Okay, we’re all going to start using it — and then let’s just keep going. This is a continuous part of our mindset forever going forward.

[38:22] The second piece — and this is something I learned when we moved away from big releases in software, which were the bane of my personal existence. You’d have to stay up all night and do them in the middle of the night. It was this long release process, inevitably you’d break a bunch of stuff, and then you’d have to spend the next day trying to fix things without having slept yet. So we moved to something called continuous deployment, where every single little incremental change in code goes out independently. And the beauty of this is that if something breaks, you can instantly roll it back.

[39:06] Now, telling software teams we were moving to continuous deployment — “Whoa, whoa, whoa. We’re going to be releasing code all day, every day. I’m never sleeping. You’ve just created a nightmare for me.”

[39:24] Yeah. I’m on all the time.

[39:26] And so what I did in that instance was say: “Let’s just do it once.” Because there was a bunch of stuff we had to build, we had to change our mindset and our process. “Let’s just do it once and see what happens.” And we did it once. And it went really well. And the team looked at each other and said, “I think we could do this again.” And then within probably a couple of months, we were deploying five times a day. So I think the real trick is: start small, have the right team, attack one problem, do it once, see the results, and then work from there.

‍

Chapter 8: Who Owns the Initiative?

‍

[40:06] So who owns the initiative then? We’re not a big massive company. We can’t have an AI department or a full tech team involved with this. Is it the process owner? What part of the organization owns these types of initiatives?

[40:23] It still comes from the executive stakeholders. Projects live and die by executive sponsorship. So if it’s meaningful to the C-suite and they explain to the organization that we’re doing this, we’re all going in the right direction, and we’re all going to reap the benefits of it even though it might be a little bumpy as we get there — it starts there.

[40:43] Where you choose to deploy your pods will probably depend on each business in itself. But we used to talk about the concept of creating super users as a nucleus on projects previously. When you’d roll something out to a big organization, you’d want to start with that super user group because they then become your advocates as things spread to the rest of the organization. So I would start with people that are change-positive. They like this type of stuff. I would not start with detractors or people that are traditionally negative or critical. Start with the teams or departments that are really positive and good with organizational change management.

[41:36] That’s awesome advice. Listen, we’re running on time. Steve, what are some last pieces of wisdom you want to tell the folks out there around starting some of this and what should they know before getting into it?

[41:50] We are at a pivotal moment in technology. I can remember two or three instances of these types of moments in my career. And so there is an arbitrage opportunity for those businesses that take advantage of it, simply by being in the game. My recommendation is: have a strategy, consult an expert, and get started. And if you’re curious, you can always reach out to Nish or myself and we can help coach you through those early steps. But the sooner you get rolling, the quicker your company is going to become AI-native.

[42:30] Yeah. Small experiments are probably the best way to go. If you think too large, you again get back to analysis paralysis and get stuck. So how do you move quickly, learn, make some mistakes as you go — but frankly, you’ll be ahead of most just by getting rolling.

[42:44] Exactly. I like that little pun there. Rolling.

[42:51] It’s been awesome, Steve. Thanks so much for joining us. Really appreciate the insights. For anybody who wants to connect with Steve or learn more, check out the Rolling Wave website — we’ll put it on the LinkedIn post — or reach out to Steve on LinkedIn.

[43:11] Yeah. And thanks for having me, Nish. This was fun.

[43:13] Amazing. Thanks so much. Take care. See you. Bye.