Drew Garner, Chief AI Officer at Smartsheet: “AI will be most transformative when it reduces friction between people, systems and data”

When you read Drew Garner’s job title for the first time, Chief AI Officer at Smartsheet, you may well wonder what on Earth that means. And the answer will no doubt vary for every Chief AI Officer you meet (we suspect that number will grow). What we can say with Drew is that he doesn’t sit in an ivory tower: alongside leading Smartsheet’s AI strategy and engineering delivery, he still builds. That means still working with clients, and still diving deep into the technical side.

If that isn’t reason enough to listen to what Drew has to say in this wide-ranging interview, part of our Conversations on AI series, consider that in his previous job, Chief Technology Officer at Accolade, he “architected a multi-petabyte data ecosystem powering AI-driven healthcare”.

One thing shone through from this particular conversation: the value of humans alongside a need to adapt – both humans and the software. “I always say that AI will be most transformative when it reduces friction between people, systems and data,” he said. Done right, breakthroughs happen.

But he has many words of caution too. “If one employee becomes four times more productive and the rest of the workflow stays broken, you’ve just moved the bottleneck,” he points out. “You haven’t solved anything.”

Read the full interview to discover his advice on how to use AI effectively, but perhaps Drew’s key insight is that we will have to rethink how we build enterprise software. That we should be building around the questions people ask rather than around a software interface.

Let’s not rush ahead, though. First, we wanted to know about the mistakes Drew felt the industry was making when it came to embracing AI.

Everyone says AI is transforming software, but where do you believe the industry is still overestimating its impact and where is it underestimating it?

We’re overestimating how much AI changes software’s fundamental purpose. AI becomes the front door, that’s the transformation. But behind that door? You still need workflows, data, governance and a source of truth. AI doesn’t eliminate these foundations but makes them even more essential. When organisations neglect this, they hand AI bad data to work with which in-turn produces bad answers at scale.

We’re also massively underestimating how AI changes the experience of using software. For decades, we forced people to climb the curve to reach value. But with AI, sophisticated capabilities suddenly become accessible without deep product expertise. Software now adapts to how people work, not the other way around. And what surprised us most was that the demand was always there.

We analysed our heaviest AI users and found that 87% had previously been locked out. The barrier wasn’t that they didn’t want to build. It was that they couldn’t reach the depth, and AI helped us eliminate it. Instead of navigating dashboards and workflows, users can now start with a question or a goal, and the system manages the coordination for the rest. Whether that’s gathering the information or bringing back an answer to act on. The expertise shifts from knowing how to use the tool to simply knowing what you want.

That’s why I always say that AI will be most transformative when it reduces friction between people, systems and data. Building a campaign or launching a product can mean hundreds of decisions and handoffs. And that’s where the real complexity lies. But that’s also where software companies win because that’s where the breakthrough happens.

Many SaaS vendors now describe themselves as “AI-powered.” What actually separates companies creating real customer value from those simply adding AI features? 

AI features are easy. What’s difficult is connecting AI to the systems and workflows that businesses depend on. That’s where value gets created. Right now, most companies are celebrating productivity gains without asking if those gains actually flow through to business results. That’s like celebrating GitHub commits instead of asking whether features actually shipped. You can measure all the activity you want, but if that activity did not move a business outcome, the metrics are meaningless.

I’m seeing companies focus on making individuals faster while missing the bigger opportunity to  help teams work better together. If one employee becomes four times more productive and the rest of the workflow stays broken, you’ve just moved the bottleneck. You haven’t solved anything.

The companies creating real value are the ones whose AI understands the actual work. A bolted-on AI feature is identical on Day 100 or Day 1. It plateaus. But when AI is connected to how organisations execute, it gets sharper every time you use it. It learns from your real workflows, your data, your decisions. The value grows instead of stalling.

What is the biggest misconception enterprise customers still have about adopting AI within business-critical software? 

The biggest misconception is that AI fixes broken foundations, it doesn’t. Instead, AI exposes and scales them.

For twenty years, weak foundations stayed contained. Only a handful of experts navigated around broken workflows and inconsistent data as part of ungoverned processes. They knew where problems were and worked around them in silos. That barrier, however, kept everyone else out and protected them from the mess they made.

Remove the barrier with AI, and that protection vanishes. Every gap in data, every disconnected workflow, every ungoverned process is now accessible to everyone. And most have no idea the gaps are there. I meet companies that bought every major AI platform but still can’t move information between departments, essentially handing the whole company a faster way to multiply the dysfunction.

The hard truth is that the same technology will either unlock your organisation or scale your chaos, depending on what’s underneath. The old foundation only had to be good enough for experts to navigate. The new one has to be good enough for anyone to reach safely. That means governing data so the answers your AI produces are trustworthy. Next is implementing clear boundaries so there’s rules on what agents can and can’t do. Most importantly, keeping human checkpoints as the norm every step of the way is a must.

The goal is to make sure that when your whole organisation can access your systems, what they reach is worth reaching, without slowing AI down.

If you were launching your company today with AI available from day one, what would you build differently? 

I would avoid building around the user interface first. I’d build around the questions people are constantly trying to answer. What’s causing delays in this project? Which customers are at risk of churn? Which initiatives will have the most impact? Most organisations have decades of workflows and expertise locked inside systems. AI finally lets you put that to work instead of letting it sit dormant. The data’s already there and the historical context is already there. You’re just building the bridge.

I’d also spend far less time asking whether software can do something and far more time asking whether people can find it. Enterprise software is full of powerful capabilities most users never touch. The simple reason is that they don’t know those features exist. AI changes that because expertise suddenly becomes accessible through conversation. An operations leader doesn’t need to know which report to run. They just ask AI the question and get the insight.

But most importantly, I’d build software that almost disappears. The best enterprise software shouldn’t require people to learn where everything lives or which dashboard to open. I don’t want users navigating the system. I want them to start with what they need and get it. For a project manager, that means asking ‘what’s going to derail us?’ and immediately getting the answer, not navigating three dashboards and stitching together data from two systems. That’s the entire job of a software company in an AI-first world.

How is AI changing the role of your customers? Are you replacing repetitive work, augmenting decision-making, or fundamentally changing how people do their jobs? 

I don’t think AI is changing what our customers do. I think it’s changing how much time they spend proving they’re doing it. Project managers and operations leaders have spent enormous amounts of time building reports, collecting updates, writing status memos and running meetings to demonstrate progress. AI makes that administrative overhead far less time intensive. That’s significant, but that’s not transforming the job itself.

What’s actually happening is the role evolves from scorekeeper to decision-maker. A scorekeeper builds the report. A decision-maker understands what the report means. A scorekeeper tracks that one project is over budget. A decision-maker recognises that the same problem is showing up across five projects and understands what pattern is actually causing it. It’s the distinction between information and insight. And that distinction actually changes how we design AI systems. 

We design around what I call the ‘expert in the middle.’ The agent analyses and recommends. The human validates the output and makes the call, especially when stakes are high.

That is the most exciting part: AI surfaces questions teams didn’t know they needed to ask. That’s where you move beyond productivity gains into actual business outcomes. When a PMO can spend 30% less time on reporting and 30% more time spotting risk patterns and understanding root causes, they can reshape strategy—that’s when the role really changes. That’s when you’ve fundamentally altered how the organisation operates and can know when AI has delivered real value.

There’s growing discussion around AI agents replacing traditional software workflows. Do you see the future as applications becoming collections of intelligent agents, or will conventional interfaces remain central? 

The industry is presenting a false choice. The future isn’t agents or applications. It’s both. Agents are great at taking action, but people still need visibility on what’s happening. You don’t run a business through a black box. We’ve already seen this play out with AI interfaces. People love being able to ask questions in natural language, but they still want visualisations and workflows when they’re making important decisions. Sometimes you want a conversation. Sometimes you want to see the data for yourself. Both have a place.

What I do think changes is where software gets accessed. Previously, every vendor wanted you to come into their application. But we’re increasingly seeing customers want the opposite. They want to use Copilot, Claude, Gemini, ChatGPT, or whatever comes next and still be able to interact with their work. We believe customers should bring whichever AI they already love into the interface of their choice. The intelligence underneath is the value add.

I also think we’re heading toward a correction in the market. Right now, everyone is building agents but at some point, enterprises are going to realise they’ve created hundreds or even thousands of them. Enterprises don’t need that many disconnected agents any more than they needed hundreds of disconnected SaaS applications. This is why orchestration and governance become more important than simply creating another agent.

We’ll look back at this period and realise we spent too much time talking about the interface and not enough time talking about the intelligence layer beneath it. The agent will change, the chat window will change, the model will change. The thing that matters is whether there is a system in place that connects the dots between them all and understands how work actually gets done.

How do you balance innovation with responsible AI? Where do you draw the line between moving quickly and ensuring customers can trust the outputs?

Innovation and responsible AI have to move together in enterprise software. If customers don’t trust what the system is doing, they won’t use it in the workflows that matter most. You can’t innovate past that.

One of the shifts we’ve made as a company is moving away from the idea that every problem should be solved with the largest, newest model available. Sometimes a large language model is the right answer. Sometimes a lighter-weight model or a deterministic workflow is more efficient and reliable. The goal is to use the right AI for the job, not the fanciest one.

We also made a deliberate choice not to treat governance as something that gets added later. Responsible AI has to be built into the architecture from day one. That’s why our framework is grounded in four commitments: data privacy and security, accountability and control, reliability and fairness, and transparency and explainability. They’re as important as the technology we build.

Trust ultimately comes from giving customers visibility and control. Customer data remains under their control and isn’t used to train third-party foundation models. AI operates within existing permission structures. Equally important, AI actions can be traced back to their source, so our customers can understand how a recommendation or action was generated. That’s the line. The goal is always to develop dependable software. We want our customers to feel confident that AI is helping them make better decisions, not making decisions on their behalf.