In kicking off Waves 2026 in Prague, Make CEO Fabian Veit says that he wants to put his customers’ AI-enabled operations inside a “glass box” — one which makes it easier for humans to understand, govern and control their future agentic organizations at enterprise scale. Because right now he says that there is a big difference between broad individual adoption and deep organizational change:

AI adoption is solved. You can now give everybody access to AI chat. People will be using it for day-to-day work, and something is happening. But at the same time, you haven't really fully transformed how work gets done and how you operate as a business. Processes somehow still run the same way as they did pre-AI. Systems and approvals have barely changed.

And the heart of the problem, at least in Veit’s telling, is simple — trust.

Not trust in the popular sense of whether AI might be untrustworthy or manipulative — or is perhaps going to become the Terminator and kill us all — but rather whether people understand enough about the workings of AI and its interaction with their operations to be sufficiently willing to take the leap of faith necessary when operational reliability is at stake.

Because while personal productivity is easy to adopt — with clear personal control and relatively low risk — the wider transformation necessary to embed probabilistic AI into ongoing business operations can feel like the exact opposite — opaque in influence, uncertain in effect and potentially existential in outcome.

And so, Veit explains, Make wants to help people build trust by helping them see and understand — by making the workings of AI transparent at every level.

In effect, to allow people to turn AI-infused operations from a black box in which you must trust your fate to an uncertain probabilistic god into a glass box in which you can understand, shape and control things for yourself.

And he explicitly ties Make's future to this vision — describing an ambition to become the visual layer through which humans will remain in control while agents do the work.

But this isn’t as straightforward as it might seem — because when thousands of agents are interacting with thousands of enterprise processes in thousands of ways, the combined activities are… complicated. And so what Veit goes on to describe wasn’t a literal offer of transparency — the ability to see the ungovernable maelstrom of technology changing everything, everywhere all at once. It is an offer of refraction — a way of organizing perspectives on the operation of an AI-enabled business that give humans leverage through meaningful abstractions.

And so to enable this the company outlined a range of developments it wants to use to make transparency workable across different parts of the organizational lifecycle — from building and governing agents through to operating them and discovering opportunities for their use.

In setting out his vision, Veit argues that agents will increasingly be the ones turning our expressions of intent into the systems required to realize them — and that natural language will therefore increasingly replace the kinds of visual modeling his company has spent years putting at the heart of its platform.

But perhaps counter-intuitively, he also argues that this will make visual models more important than ever — because they will become a critical medium through which humans will understand and approve change as its speed, complexity and distribution become too great for us to follow every detail. In that sense, visual models become a form of refraction — representing agents’ proposed changes in ways that enable us to control intent without having to control implementation.

Veit even goes so far as to suggest that the days of drag-and-drop development are largely numbered — a pretty interesting prediction from the CEO of a company whose visual scenario builder has been central to its identity for more than a decade.

To back up this aspect of its narrative, Make explains how its AI assistant, Maia, can work with users to plan and design solutions alongside them, absorbing requirements, discussing options and setting out its intended approach for approval before proceeding. And while Veit might be — perhaps a little slyly — suggesting that drag-and-drop has had its day, it is telling that these discussions with Maia can also involve diagrams and screenshots.

Continuing with its longstanding integration stance, Make is also opening its environment to external assistants such as Claude and ChatGPT through MCP, allowing users to work with whichever agents they prefer. But regardless of which agent does the building, the resulting solutions remain visible on Make's canvas, where people can inspect, question and change what has been constructed.

Looking beyond the shaping of intended behavior to what agents can actually do at runtime, Make's Head of Product for Governance, Anastasia Horňáková, describes the opacity surrounding many current deployments — from shared team accounts and overly permissive admin access to people being free to use whichever AI models they choose. In such environments, it becomes difficult even to establish who is doing what, under whose authority, and for what purpose.

And while such permissiveness might have been less than ideal — but manageable — when operational activity was largely human-led or governed by deterministic systems, Horňáková argues that it becomes untenable in an environment where agents will be doing things we don't know in advance, at a scale far beyond what we're used to. But she also argues that this doesn't mean we can't control our AI-enabled operating models — it just means we need to be more intentional about the tools we make available to agents, the actions those tools permit, and the authority we delegate to them.

And if we can make those boundaries explicit — by refracting them through clearly defined permissions, restricted tool access and visibility into who is acting under whose authority — then it becomes much easier to trust the resulting operations. Because discretion that would otherwise be buried in the behavior of individual agents becomes something we can understand and constrain.

Make's proposed approach therefore draws on its existing control structures. Agents using MCP inherit their human owners' roles and permissions, while private spaces isolate sensitive resources and MCP toolboxes restrict agents to specific connections, scenarios, knowledge and actions. And where those preconfigured boundaries are insufficient — because an action requires a judgment that the organization isn't willing to delegate — Make's forthcoming human-approval capabilities allow workflows to pause and return the decision to a person before proceeding.

Moving beyond the visibility of individual agents, processes and integrations, another Make Head of Product, Peter Trenkner, makes what seemed to me to be one of the most consequential observations in the whole ‘glass box’ conceit — namely that what happens within individual systems tells us little about how they relate to each other or contribute to the collective operation of the enterprise. Because in effect thousands of individually legible assets don't necessarily add up to a collectively legible operating model.

And so his argument is that, to truly see the enterprise as a glass box, we need to understand how it works as a whole — not just its constituent components, but the relationships between them and how work flows, for good or ill, across the resulting landscape. And that means understanding both how the components are connected and how the operating model they constitute actually behaves — otherwise we will struggle to have any real confidence in its operation or to decide where we need to intervene.

And this is where Make's existing landscape visualization technology, Grid, came to the fore by providing a visual representation of the digital assets and relationships within its customers' Make environments — one which the company is extending to include agents and assets running on other platforms. In particular, Make is planning to give agents — both its own Maia agent and external alternatives — access to Grid's underlying structures, so that people and agents can use a shared understanding of the operating model to interpret intent, examine the implications of change, and agree proposed changes.

Complementing these structural enhancements, Trenkner also outlined improvements to trace business activity across connected executions and evaluate agent performance, while demonstrating how cost information could be visualized on the Grid within the context of the wider operating model.

The final pillar of Make's presentation is a capability called ‘Discover’, which aims to scan a customer’s operations in search of processes or tasks with the potential to be improved through Make-based automations and agents.

Unsurprisingly, the foundation for this emerging capability uses the well-established process intelligence technology of Make’s parent company Celonis. By combining both process mining and task mining — one to understand how processes actually operate and the other to look for manual tasks — the goal is to enable customers to cast the net wider in search of opportunities to improve their operations through new or expanded Make-based automation.

Given the differing architectural lineages of the two companies, it felt conceptually neat but a little less resolved than the rest of the narrative at this stage — but the emerging integration itself was an interesting signal of how things might develop.

At Waves this year there was a clear continuity in direction — both from the perspective of Make’s longstanding focus on visualization and also in terms of the increasingly central value of its Grid technology.

One new element that stands out, however, is the introduction of Celonis’s process intelligence technology into the Make offering. I could still see the seams where the two products were being smooshed together, but the portfolio logic is difficult to fault — and in fact it's simply surprising that it hasn't happened sooner. But having been on the acquired side of such a relationship in the past myself, I don't envy them the work necessary to reconcile two different ways of structuring data, two different ways of understanding processes, and two different ways of visualizing work — never mind the implications for each company’s respective offerings.

But beyond that family housekeeping, I liked the way in which Make's focus on visualization was cohering into its ‘glass box’ conceit — something that hints at a growing shift from the visualization of individual development activities to the visualization of collective operations. And having recently spoken to CTO Patrik Šimek about Grid, I felt as though I can see a pretty clear throughline between Veit's ambition and what Šimek had previously described — with Grid’s ability to aid the exploration and understanding of the organizational landscape feeling somewhat like a literal manifestation of the glass box. The preview of MakeOS, a new product trailed by Šimek during the keynote, in which agents will take on a broader and increasingly proactive role, also suggests that this direction is likely to intensify.

Which meant that the keynote's organization around Discover, Build, Govern and Observe struck a slightly flat note for me — because having explicitly described a future in which agents do the work and humans remain in control, Make then seems to fall back on the familiar functional compartments of a traditional software lifecycle. Compartments that still largely described the proposition through the lens of human-legible lifecycle steps rather than the continuous steering of agents the glass box vision seems to promise.

To be fair, maybe that’s what happens when you need to talk to a room full of builders who still need familiar hooks to understand where the platform is going. And perhaps I'm also getting ahead of Make here myself to some degree, because I'm not entirely sure it shares my interpretation of the value of Grid to its future — or indeed how much of that future the company can really deliver.

But when Veit’s own vision increasingly leaves the platform — and whatever internal lifecycle it supports — as the plaything of agents, then it feels like the real value inevitably shifts towards surfaces, like Grid, that maximize shared organizational understanding and control.

Which is why, honestly, it feels like pretty much the most glass box thing I've seen.