Lately, people have started asking me for my advice on how to accelerate enterprise AI adoption. Probably because I literally can't stop talking about all the great things happening at Tines since we gave every employee freedom to build with AI.
My advice, of course, isn't perfect. We rolled out AI coding capabilities to all our employees in July and I'm writing this in November, so our company is still relatively early in our AI transformation.
But as I think about what's worked for us, and chat to customers who are on a similar path, some really strong patterns start to emerge. Things like ownership, controls, knowledge sharing, and how you measure success are proving to be really important.
Our approach to AI transformation is not something other leaders can easily copy outright. It's influenced by the technology we use — our own product, Tines 3B — and by our company culture, which for the last eight years has been heavily focused on automation.
But even though our teams have always been keen adopters of new technology, there’s been a huge shift since we rolled out 3B to all of our employees. Something fundamental changes in a company's culture when building with AI suddenly becomes safe and easy — when people no longer need to ask for permission, they can start building and testing just as soon as the idea appears in their head.
So today, I'm sharing learnings from our recent 3B rollout, and from the many valuable customer conversations we've had along the way.
Enterprise AI transformation: the Dos
1. Find an owner for AI enablement from day one
This sounds obvious, but it's surprisingly easy to overlook. Somebody at your company needs to "own" AI transformation, and be accountable for its success.
One customer recently told us they had "lost traction" because nobody was responsible for driving enablement. Another took six months to find the right internal AI champion. In both cases, teams only made real progress once someone was in the driver's seat.
At Tines, we appointed one of our longest-standing employees, Martin Moroney, as our Head of AI Transformation. We deliberately chose someone with widespread respect across the company, and the kind of person who could spark enthusiasm and unblock projects fast.
2. Put controls in place before opening the floodgates
Impactful building can only happen when teams have access to the essential systems and data they need — which means the right controls and governance have to be in place from the start.
This is one of the biggest challenges we hear about from customers, and a big reason why a single, complete environment for AI building, rather than a patchwork of tools for things like AI harnessing and observability, is resonating so strongly.
You can't expect teams like finance, HR, and marketing to suddenly understand the ins and outs of corporate security. These controls need to be built into the environment itself, so those teams don't have to think about them at all.
That's how it works at Tines. Because everything runs through 3B, our security and IT teams set the controls: which connectors exist, who can access what, and what gets published where. Everyone else just focuses on building.
3. Get your executives building (and sharing their progress)
Senior execs aren't always the most hands-on when it comes to emerging technology. But when leaders build with AI in a way the whole company can see, they give everyone else permission to experiment.
These days, I spend most of my day in 3B, and I know I'm not alone; one founder I spoke to told us he's "building a ton" of AI-assisted software himself.
When senior leaders share what they've built, whether on an all-hands call or in a Slack message, it sends a much stronger signal than a company-wide email ever could.
Execs also need to reinforce the organization's AI strategy: what teams should focus on, how they collaborate, and where the guardrails are. If they don't, teams will build anyway. In this way, executives face a choice — decide the direction their teams go in, or have this decided for them.
4. Start with small but meaningful use cases
My advice to teams just getting started with AI-assisted building is this: find a task that frustrates a whole team every week. Something repetitive, relatively low risk, and easy to recognize when it's been improved, but that benefits the team as a whole, not just the person who builds it.
One senior director of AI enablement recently told me, "We don't have to start with the most complex and highest value thing. Just find something that can help create a pathway to more complex things."
Individual productivity hacks are a good sign that people are experimenting. But if you want momentum, look for the task that's bugging ten people, not one.
5. Let people closest to the problem build the solution
At Tines, we've proven this again and again: the person closest to the problem is usually the best person to fix it.
When most of our automation ran through our original product, Tines Stories, more technical team members would sometimes help colleagues in teams like finance, HR, marketing, and legal build things. The results didn’t always hold up.
When people deeply understand a process, including its edge cases, they can build exactly the right app, agent or automation, not just a serviceable one.
6. Give people time to experiment
One CIO told us he noticed some of the most impactful building was happening on the side, squeezed in around people's day jobs. His organization has now made it a company-wide mandate: building with AI counts as real work, with dedicated time carved out for it, not something people have to find room for on their own.
You can't tell people to embrace AI and then expect them to squeeze it into an already full week. If it matters, it needs to show up on the calendar.
7. Find champions in every team
The old Center of Excellence model — a single group setting the rules, owning the expertise, and making the calls on how AI gets used — made sense when adoption was new and contained. It doesn't scale once people across the business are building.
This group’s job isn't to own AI excellence; it's to create the conditions for it to happen across the whole company.
In practice, that means finding people in every team who can answer the first few questions, spot opportunities, and help colleagues get unstuck.
One organization we spoke to set up a dedicated Slack channel for AI champions, where builders post what they've made, ask questions, and help each other troubleshoot. Another recruited hackathon participants as ongoing champions, giving them continued access in exchange for helping others build.
8. Give people a reason to build together
Hackathons, build weeks, and regular demo sessions are great ways to create momentum.
One company I chatted to ran an "AI surge week" with temporarily increased token limits. Another runs monthly demo days where employees show off everything from time-saving automations to full applications.
We did something similar at Tines, running a company-wide 3B hackathon before launch. Many teams now run their own quarterly hackathons.
The important thing is not to treat these as one-off events. Give people regular opportunities to experiment, share what they've learned, and see what their colleagues are doing.
9. Make it easy to see what other people are building
One IT director told us they'd discovered seven people had independently built tools to solve exactly the same problem.
That's not necessarily a failure — duplication is often a sign of healthy experimentation. But seven versions of the same tool is also a good argument for making all builds visible to IT, rather than relying on people to stumble across each other's work.
At Tines, 3B monitors for similar patterns across builds — the same connectors, the same API calls — and flags potential duplicates, helping us spot opportunities for improvement and consolidation.
10. Plan for what happens once a build is done
What happens when something an employee built with AI becomes so useful that a whole team, or multiple teams, now rely on it?
Who's responsible for that build? Who maintains it today? Who maintains it if the original builder moves on? As your organization's AI adoption matures, your people will probably start building mission-critical software, and these will become really important questions to answer.
At Tines, this is all managed within our 3B environment, where every build has an owner by default. That means our IT and security teams can see who built it, what it's doing, and what it connects to, and get alerted when something breaks or stops working as it should.
11. Oversight should scale with risk
Not every organization has oversight built in by default. Several leaders we've spoken to have taken a more strategic approach: assign oversight based on risk, rather than applying the same process to everything.
For example, some organizations I've spoken to let people experiment freely with personal workflows, while introducing an approval step only for builds that connect to core enterprise systems or get shared across an entire team.
Whatever process you land on, make sure the high-risk items always have oversight.
12. Be clear about which tool solves which problem
Most employees already have access to some AI tools. Adding another one without explaining how it fits alongside them just creates confusion.
One AI enablement lead gave me a useful rule of thumb: if it's something you ask for once and then move on from, a chat assistant or bolted-on AI tool is probably fine. But if it needs to keep running reliably in the background, without you, it belongs in an environment with the appropriate level of governance.
Without that distinction, people can misunderstand or underuse the tools available to them.
13. Remember not everyone is comfortable with these tools
Non-technical employees have told us they found AI coding tools "scary" or "daunting" before they actually started using them.
Meanwhile, technical teams like engineers can be skeptical of platforms that give everyone the freedom to build with AI.
Both groups need help understanding what these tools are for, and how they benefit their specific role. It's another reason why knowledge sharing, and surfacing inspiring examples, is so important.
14. Measure what's actually changing
Start tracking the impact of your teams’ software builds from day one.
Impact shows up in different ways, and it's worth tracking all of them:
Saving money — software you no longer need to buy, or vendor costs you've replaced
Protecting or growing revenue — deals closed faster, churn caught earlier, fewer dropped leads
Increasing team capacity or speed — the same team handling more volume, or getting through work faster
Increasing speed of innovation — ideas going from concept to production in days, not months
Improving quality of outcomes — fewer errors, better compliance, more consistent results
Sharing these wins is just as important. People are far more likely to get involved when they can see what their colleagues are achieving.
Enterprise AI transformation: the Don’ts
1. Don't tell people to use AI without giving them somewhere to build
One IT leader told me that his organization encouraged all its employees to build agents on its core AI platform. Requests flooded in, a backlog formed that was "a mile long," and the IT team had to call a halt.
If you encourage people to experiment but don't give them a safe, easy way to do it, momentum will stall — or worse, people will find their own unsanctioned ways to build.
2. Don't make one team responsible for building everything
One CISO told me that their biggest bottleneck was a single AI analyst who was "the only one who can build stuff."
This is what happens when you separate the person who understands a problem from the ability to solve it. Appointed teams can set standards and help with complex use cases, but if every request has to go through them, you'll quickly create a queue nobody wants to wait in.
3. Don't let one group own AI excellence
We touched on this earlier: centralizing all the expertise in one group doesn't scale once the whole business is building. But that doesn't mean a central group has no role at all.
They should still set the rules — security, access, what's safe to connect to, how things get deployed. What they shouldn't do is own the expertise or decide what gets built. Their job is how building happens safely, not what's worth building in the first place.
At Tines, we centralized the platform, which decentralized excellence. Builders and champions across the business carry that forward themselves, through office hours, shared workflows, and both formal and informal knowledge-sharing.
4. Make the safe option the fast option
A security director recently told me about an employee who bought their own cloud account because getting access through the approved route was too slow. A friend at another organization described their AI approval process as "a nightmare."
These are warning signs. If the official path takes weeks, people will find another one — usually an ungoverned one. Organizations need to make the safe option the easy option.
5. Before you say no, find an alternative
One IT team we work closely with never just says no when an employee asks for a new tool — they look for an alternative to offer instead. I really admire that approach. It assumes people are smart and good at their jobs, and just need help finding the right way to do them.
Of course, not every team is in a position to say yes as often as we are at Tines. Security, compliance, and budget constraints look different for every organization, and sometimes the answer genuinely is no. But it's worth looking back at your team's recent "nos" and asking: could we have offered something more than a flat refusal?
6. Don't mistake a policy document for an adoption strategy
Lots of organizations have an AI policy and an AI council. Far fewer have figured out how to get people building business-critical software at scale.
Policies matter, but they don't create adoption on their own. People need the right tools and support to make real transformation happen.
7. Ensure builders know what happens after the build
One security manager described what happens after someone creates their first workflow: "The person who built the thing has no idea what to do next."
People need to know what the next step actually is: whether that's a review process, a publishing step, or getting sign-off from an approver. Make that step obvious before someone finishes their first build, not after.
8. Don't assume every team wants the same thing
Engineering teams already have their own environments for running, testing, reviewing, and securing what they build. But teams like finance, marketing, and HR aren't even aware these things are needed in the first place. If governance and observability aren't built in by default, and you're left explaining concepts like AI harnesses to people who have no framework for understanding them, you're fighting an uphill battle.
9. Don't default to agents when a deterministic workflow will do
For builders, there's a tendency to jump straight to agents because they're exciting. But sometimes a deterministic workflow is exactly what's needed.
At Tines, our teams have built hundreds of business-critical software products since we rolled out 3B, but around 77% of them run only on deterministic steps, keeping costs down and repeatability high.
10. Don't measure adoption by licenses or logins
Giving someone a license doesn't mean they're using AI. Logging in doesn't mean they're getting value from it.
When you're evaluating AI adoption, look at what's actually being built, what's making it into production, and what impact it's having. For a benchmark: at Tines, 90% of employees are actively building with AI since we rolled out 3B, not just logging in, but shipping things that get used.
11. Don't make it about reducing headcount
One consulting director told me the CFOs they work with rarely talk about cutting people. What they actually want is for teams to stop waiting months to solve mission-critical problems.
That's a much better way to frame the opportunity. AI should help unblock people, not replace them, giving them the ability to fix what's broken today, instead of waiting on a roadmap, a budget cycle, or someone else's availability.
How to tell when AI transformation is working
Borrowing best practices from other companies can be helpful. But you also need to read the signals of your own organization to measure whether any of this is actually working. Here are some signs worth watching for.
You'll know things are moving in the right direction when:
People outside technical teams like engineering, IT, and security are building things without being asked
Teams are reusing and improving each other's work rather than starting from scratch
Requests to IT shift from "can you build this for me?" to "can I have access to this system or data?"
Ideas are making it into production in days, not months
Fewer tools are appearing on personal accounts and laptops
Leaders can easily answer two questions: What are we getting out of AI, and are we confident in how we're managing the risks?
