Chris Perry, Founder and CEO of Andus Labs – Interview Series

Chris Perry, Founder and CEO of Andus Labs, is a communications, digital strategy, and innovation executive with nearly three decades of experience helping organizations navigate major shifts in technology and business. Before founding Andus Labs in 2025, Perry spent more than 22 years at Weber Shandwick, where he held a succession of senior leadership roles including Chairman of Futures, Chief Innovation Officer, President of Digital Communications, EVP of Digital Strategy and Operations, and SVP. Earlier in his career, he worked in technology communications at General Motors and served as Vice President at Edelman. His career has focused on digital transformation, emerging technologies, organizational change, and the evolving relationship between technology and the people responsible for putting it to work.

Andus Labs helps enterprises translate AI capabilities into practical organizational outcomes by focusing on the human systems surrounding the technology. The company describes its mission as building the “Human OS for AI”—an operating layer designed to orchestrate work across people and AI agents—based on the premise that AI investments often fall short when organizational readiness, workflows, and adoption lag behind the technology itself. Its work spans AI work orchestration programs, products intended to reinforce and compound organizational learning, and a broader network of practitioners, artists, and humanists exploring how humans and AI can work together effectively.

You spent more than two decades at Weber Shandwick, progressing from senior digital roles to Chief Innovation Officer and Chairman, Futures, before founding Andus Labs in 2025. What did you observe inside large organizations during that period that convinced you the biggest barrier to AI transformation would be the human and organizational system around the technology rather than the technology itself?

I learned it first at my dad’s graphic arts studio in Detroit, where I started out, before Weber Shandwick. Desktop publishing moved the work his typesetters and illustrators did by hand into software. It broke his business model, and he didn’t see the shift until it was too late.

At Weber Shandwick I worked with some of the world’s largest companies through new tech waves: social media, blogs, video, mobile, data. Each time, the pattern was the same. The technology arrived. People felt the change and started using it on their own. Leadership and the structure of the company moved last.

In those earlier waves, companies paid for that lag by losing ground, but people still did the work. AI does to knowledge work what desktop publishing did to my dad’s studio. The way people do the work gets built into the software.

Business schools don’t teach you how to run a company where people and agents share the work. Most companies bolt AI on and get the old work done faster. The real return comes when leaders set their people up to rebuild the work with AI. The work changes shape instead of just getting faster.

We named it Andus Labs, as in “and us,” because we bet the payoff would come down to people.

Andus Labs argues that enterprises can have advanced AI capabilities while still lacking the “human readiness” required to capture value from them. What separates a company that has successfully deployed AI from one that has actually transformed how work gets done?

I sat with a company reporting 80% AI adoption and something closer to 5% change in how work got done. Only the adoption number made the board deck. The second number wasn’t being hidden. The dashboard just had no place for it.

Deployment is easy to measure. Licenses assigned, agents deployed, logins. That’s activity, not impact. We ask clients what the company can do now that it couldn’t do before. Most of the time the honest answer is the same reports, decks and analysis, only faster.

Companies that are transforming measure something else. Did the pilot produce enough value to be worth scaling? Can we put something in front of the board that we couldn’t have produced a year ago? Is AI taking weight off our most valuable people so they can do the work only they can do? None of those are as easy to count as seats, but they’re the measures that tell you whether anything moved.

Those questions also need an owner. Ask who that is inside most companies and you get a list. The CIO owns the platforms. A center of excellence owns the pilots. HR owns the training. The impact on the business sits on nobody’s list.

One company can tell you how many people have access. The other can tell you what changed and who answers for it.

The Ground Truth Index now draws from a library of 300 documented patterns observed through engagements and conversations with leaders across 595 organizations, with five independent analyst agents scoring each pattern before researchers review and adjudicate the final rankings. How has this methodology evolved, where does human judgment enter the process, and how do you prevent the Index from simply reinforcing assumptions already present in your field observations?

The method has changed as the library has grown. Every pattern starts with what we hear and see in conversations with business leaders. We check each new signal against the library, test it against field observation and outside research, and decide whether it’s genuinely new, strengthens a pattern we already track, or belongs elsewhere.

AI helps us assess the whole library every quarter, which a research team couldn’t do by hand. Agents read every pattern through a different professional lens and score it on the same five dimensions. They work blind, so no agent can influence another. We take the median so a single outlier can’t swing the score, and the top 50 go to human review.

Human judgment runs through all of it. Our research team approves every pattern that enters the library, reviews the rankings and decides the final 25. We look hardest where the agents disagree, and where media attention conflicts with what we’re hearing in the field. We can move a score, but only on documented observation: who raised it, how often we heard it, and whether it held up across organizations. A hunch doesn’t move a score.

The new Ground Truth Index ranks “Empty Chairs” as the number-one barrier to enterprise AI, describing organizations that give AI systems influence over decisions such as hiring, pricing, and risk without clearly identifying who is accountable or has the authority to overturn those decisions. As AI moves from recommending actions to actually shaping business outcomes, what should meaningful human accountability look like in practice?

The more functions AI touches, the more decisions it makes on the company’s behalf. Accountability becomes a management question before it’s a technology one.

Responsibility usually sits somewhere between whoever bought the AI system and whoever uses it, and neither one owns it. The gap stays invisible until a decision gets challenged. That’s the empty chair.

Accountability means a named leader who sees what the system produces and can explain the outcome, correct it or stop the system. That takes authority over a process that usually crosses several departments.

We ask leaders to pick one consequential decision their AI helped make this quarter and tell us who’s accountable for it. When the answer isn’t clear, the company has changed how decisions get made without settling who answers for them.

“Learning While Drowning” ranks second in the new Index, highlighting companies that mandate AI training while leaving employees’ existing workloads unchanged. At the same time, “Optimization Temptation” shows how easily companies can quantify headcount savings compared with the harder-to-measure value of redeploying people into higher-value work. Are companies fundamentally underestimating the organizational investment required to make their workforce more productive with AI rather than simply using AI to reduce costs?

Yes. The savings from eliminating a role are easy to put on a spreadsheet. The value of that person’s knowledge, relationships and judgment is much harder to capture. The savings show up this quarter. The losses can build for years.

The assumption behind the cut is often wrong, too. AI makes today’s work more efficient, and it also turns up new work that needs doing. What we’re seeing is teams rebalancing, with people moving into different jobs rather than whole functions getting cut.

A training mandate is a workload decision. Companies pile AI training on top of existing work, then read low completion as a motivation problem. The investment they skip is the one that never makes the budget.  Deciding what work will stop, shrink or move to make room, and naming who signed off on it. Learning takes time. Applying it takes more, because people have to test unfamiliar tools, check the output and change how they’ve always worked.

The answer isn’t another item on the list. Put a team on a live project and the learning happens inside the work.

The Index also identifies “Devaluation Anxiety,” reflecting concerns about what happens to employees when AI becomes capable of performing increasingly valuable parts of their jobs. How much of what executives label as resistance to AI is actually rational anxiety about career value, and what should leaders do differently when introducing these systems?

Most of it is rational. Employees hear about AI efficiency in one announcement and job cuts in the next. They draw their own conclusions.

Leadership treats those as separate messages. Teams don’t. They finish the training and hold back the experimentation that would have made it useful.

Reassurance won’t fix that. Leaders have to say what’s been decided about jobs, what’s still uncertain and what support people can expect.

The same goes for the payoff. Companies spell out what the business expects from AI and leave the employee’s side blank. If someone gets faster, what do they get out of it? Room for better work, a path to advancement, time back? That belongs in the rollout plan.

Executives are asking their teams to work differently with AI. The teams are waiting to see the executives do it first.

Another pattern you identify is “Active Inertia,” where organizations use AI to accelerate existing processes without fundamentally changing them. What does genuine AI-native workflow redesign look like, and can you give an example of the difference between making an old process faster and rebuilding the process around what AI now makes possible?

Most companies point AI at whatever is easiest to automate, which is often work that shouldn’t exist in the first place. Reports come out faster. Approvals move at the pace they always did. That’s Active Inertia.

Redesign decides up front what the AI models do and what the team carries, then building for that instead of inheriting the old shape.

Take a weekly report. The faster version has AI draft it, and the same people review the same document on the same schedule. The rebuilt version asks what the report was for. If it exists so someone can catch problems early, the system can flag a problem the moment it happens, and the report goes away.

You have written that “value equals capability times absorption” and argued for systems rather than simply software. As frontier models increasingly become commoditized and available to everyone, could an organization’s ability to absorb and reorganize around AI become a more durable competitive advantage than access to the models themselves?

Yes. The deploying models is easy compared to changing work around them. They arrive with extraordinary power, get cheaper every quarter. The same models are available to anyone. What you can’t buy is unique context within the company and teams ready to use them to their full potential.

What we call “receiving capacity” is an old idea. It’s a team’s ability to take in something new and put it to work, and it’s made of decision rights, incentives, workflow, job design and enough trust that people tell you what went wrong. None of it can be rented.

I describe value as capability times absorption because the two multiply. They don’t add. A brilliant model dropped into an organization that can’t take it in is a flawless part bolted to a failed seal. The capability is real. The return is still zero.

Agentic AI introduces another layer of complexity because systems can increasingly take actions, use tools, and pursue objectives rather than simply generate responses. Where should enterprises draw the line between agent autonomy and human decision-making, and what new governance structures will be required as agents receive greater authority?

Start with one question: who answers when the agent gets it wrong?

Draw the line between what the agent decides and what a person decides before anything goes live. Most companies find that line after something has already broken.

Where a mistake is cheap to undo and shows up fast, let the agent run. When the decision changes what someone earns, whether they keep their job or where they stand legally, a person makes the call and the agent prepares it.

The instinct is to centralize. Stand up a committee, write a framework, wait until it feels manageable. By the time guidance from the top reaches the work, it’s out of date, and people who can’t get an answer start using tools the company never approved.

Governance has to sit close to the work. Teams need to know what each agent can decide, who owns it and how to stop it. And they need a record of what it did.

The org chart was never built for agents that work across functions, so no single function owns the failure. Settle that before deployment, not after.

Looking ahead, if the biggest constraints on enterprise AI increasingly become organizational rather than technical, how do you expect companies themselves to change? Which parts of today’s org chart, management structure, job design, and decision-making processes are least likely to survive the transition to organizations built around humans and AI agents working together?

The functional org chart is the least likely to survive. Most companies still run by function, with work passed from one department to the next and a sign-off at every handoff. AI produces work in minutes while the approvals around it take weeks. Leaders are being asked to move faster inside systems built to slow things down.

That chart was built for people handing work to people, not for decisions shared between people and agents.
A committee won’t sort this out.

What replaces the chart looks like small teams organized around a problem. A leader names a problem worth solving, decides what has to be true to solve it, and puts a team of people and agents on it. That team gets its own funding and protection, because it can’t run the way the rest of the business does. The people on it make the judgment calls and own what the agents produce.

Electric motors reached American factories in the 1880s, but the productivity gains didn’t show up until the 1920s. At first, plants put one big motor where the steam engine had been, drove the same line shafts and left every machine where it was. The gains came when they gave each machine its own motor and laid the floor out around the work.

The rearrangement was the revolution.

Thank you for the great interview, readers who wish to learn more should visit Andus Labs or learn more at the Ground Truth Index.