Tines co-founder and COO Thomas Kinsella recently joined Peter High on the Technovation podcast to talk about the evolution of Tines, the rise of "wild code," and why agents aren’t always the right tools for the job.
The conversation covered a lot of ground — from the original problem that inspired Tines, to how teams should think about combining AI agents, deterministic automation, and humans in a single workflow.
Here, we’ll share some of the highlights.
Tines started as the platform the founders wished they had
Before founding Tines, Thomas spent close to a decade in information security, including roles at eBay and DocuSign. The problem he kept running into was a familiar one to security teams everywhere: too much work, not enough staff, and a sprawling set of tools that all needed to be connected.
We tested about 10 different automation tools, and we frankly just didn't like them. We thought they were way too complex, and we said we could do better.
That frustration led Thomas and co-founder Eoin Hinchy to start Tines in 2018 — not with the goal of building a generational company, but of building the tool they wished existed. Nearly eight years later, that mission hasn't changed: helping people build, run, and monitor their most important workflows.
From low-code to AI-native
Since launching in 2018, Tines Stories has been serving security and IT teams at organizations like Mars, Snowflake and Reddit. But as Thomas explains, the last 18 months changed the equation entirely.
People are using Claude Code, people are using Codex. And now we've got another challenge, which we call wild code.
Wild code — invisible, ungoverned agents, scripts, and automations being built across an enterprise with no central visibility or ownership — is, in Thomas's words, a problem "every enterprise is experiencing."
And it's why Tines built Tines 3B, a platform that gives teams the freedom to build with AI and IT the control to govern that work. Launched in July 2026, Tines 3B already counts Headspace and Fin (formerly Intercom) among its customers.
Democratizing building without democratizing risk
Thomas also shares how the team at Tines think about balancing access to AI coding tools with governance and control. It’s not about restricting access, or handing it out unconditionally — it's about giving them a safe path in which to build.
You make it so that the most productive path to using AI is the governed path. That's the path that has visibility, and that's the path that has control.
This is the same logic that's long guided platform engineering teams: rather than making people go find their own credentials or piece together their own access, you provide a governed, ready-to-use path that's genuinely faster than going around it. Make the secure way the fast way, and adoption follows naturally.
When to use AI, when to use automation, and when to employ a human
Thomas also shares his take on AI agents, and why they aren't always the right tool for the job.
There's often a much cheaper, much faster, reliable, deterministic method of doing things, rather than a probabilistic, expensive method of using an agent.
His framework for deciding which to use — deterministic logic, an AI agent, or a human — comes down to a few simple questions:
How predictable is the task?
What's your tolerance for error?
How fast does it need to run?
How much are you willing to spend?
Highly predictable, repeatable tasks call for deterministic automation. Fuzzy problems with a rough, known outcome are where agents shine. And judgment calls — the kind with real business context and consequences — still need a human.
Just as importantly, Thomas argues that workflows aren't static. They evolve as you learn more about them.
As you run them several times, so your logic should evolve... That's what's really exciting about working with some of these organizations.
Turning tech debt into a shared asset
A recurring theme in the conversation is the idea that "building" was never really the hard part — running and monitoring workflows reliably at scale is. Thomas describes how Tines 3B's monitoring layer can detect when a workflow's output doesn't match what's expected (a 404 instead of an approval, for instance), automatically investigate what happened, and propose a fix.
It can tell you, 'Hey, it looks like this workflow broke. Do you want to ship a fix into production?'
He also talks about turning what would traditionally become tech debt — ten different people building ten different versions of the same workflow — into a shared, reusable service across the business, rather than a fragmented mess nobody can account for.
Final thoughts
These are just a few highlights from Thomas's chat with Peter. Watch the full episode above to hear how enterprise customers differ from cloud-native ones, why "shadow AI" is really just the latest version of "shadow IT," and lots more.