.svg)
In the Privacy Soapbox, we give privacy professionals, guest writers, and opinionated industry members the stage to share their unique points of view, stories, and insights about data privacy. Authors contribute to these articles in their personal capacity. The views expressed are their own and do not necessarily represent the views of Didomi.
Do you have something to share and want to take over the Privacy Soapbox? Get in touch at blog(at)didomi.io
Last quarter, I shared our renewed vision at Didomi: helping organizations move beyond “dead-end data” and build better relationships with their customers through high-quality, consented first-party data.
As we enter the last stretch of 2026, that ambition is raising some fundamental questions. What happens to our products when customers increasingly delegate their work to AI? What does privacy mean when agents exchange personal data on our behalf? And how can brands preserve a direct relationship with customers who no longer visit their websites?
These questions have been on my mind throughout Q3. They are closely connected, and their answers will shape both what we build at Didomi and the digital ecosystem our clients operate in.
Didomi's next user may be an agent
For years, SaaS companies have built products around a familiar assumption: users log in, navigate an interface, configure settings, and click through workflows to get something done.
I believe we need to stop treating that as the default.
Our clients want their privacy operations to work. Learning our interfaces and spending hours configuring them is a means to that end. As AI becomes capable of doing more of this work, their expectations will change radically. Ours should too.
That is the thinking behind our recent initiative around the Model Context Protocol, or MCP: opening a path for AI assistants to interact directly with Didomi’s capabilities, within the permissions our customers give them.
The broader ambition goes much further. I imagine a product that assists users throughout their work, understands their context, and proactively recommends what needs attention.
A privacy team should be able to ask where its implementation is inconsistent, understand the consequences, and review a proposed correction. A marketing team should be able to understand which data it can use for a particular purpose without manually reconstructing the answer across several systems. Over time, more of that work should happen automatically, with appropriate oversight and a clear record of what changed.
This also forces us to ask an uncomfortable but necessary question: why will clients continue using Didomi if an AI agent becomes their main interface?
My answer is that our value must sit in the trusted infrastructure behind those interactions. Organizations still need a reliable source of truth for the privacy signals attached to their customer data: choices, preferences, permissions, restrictions, and the history needed to understand them.
They also need those signals to be applied consistently across their systems.
That is the role I see for Didomi: a trusted privacy partner that both people and agents can rely on. An interface can change very quickly. Maintaining an accurate understanding of what an organization can do with personal data, and making that understanding operational, is a much deeper responsibility.
Where privacy meets AI governance
The same transformation is happening within our clients’ organizations.
As companies connect AI models, internal tools, and agents, personal data can move through more systems faster, with fewer people reviewing each step. In an increasingly agent-to-agent environment, one request might involve several services before an answer reaches the user.
How do we keep the original privacy conditions attached to that data throughout the journey?
This is where I see an important role for Didomi within AI governance, and where the idea of a privacy-focused AI Gateway becomes particularly interesting.
There are three needs I believe we should address.
First, identify personal and sensitive data at the point of collection. Organizations need to understand what is entering a workflow before it is copied into prompts, shared with another service, or made available to an agent. Discovering the problem after the data has spread is already too late.
Second, establish privacy lineage. By this, I mean maintaining a traceable connection between data and the conditions governing its use: where it came from, why it was collected, which choices the person made, and how those conditions apply as the data moves. A preference change or withdrawal should trigger action wherever it matters.
Third, govern access and exposure. An agent should receive only the data necessary for its task, under the appropriate permissions. Our privacy governance layer should help enforce that through data minimization and privacy-enhancing technologies, including masking or pseudonymization where appropriate.
Consider an assistant handling a customer request. It may need an order reference and delivery status. It should not automatically receive the customer’s full profile, unrelated purchase history, and marketing preferences simply because those fields happen to be available.
This is the kind of practical problem I want us to help solve.
Our experience with consent, preferences, and server-side data flows gives us a foundation to build on. The next step is to explore how that expertise can help organizations govern data use across AI workflows, with decisions they can explain and evidence they can inspect.
Are personal agents becoming a new generation of adblockers?
A year ago, I wrote about AI agents becoming intermediaries between people and the web. This quarter brought more concrete examples of that shift.
Meta introduced Muse in September, presenting a personal agent that can work across apps, browse, and carry out tasks on a person’s behalf. Grok Bot, introduced in August, similarly puts agents to work across tools and websites. These products serve different needs, but both illustrate the move from asking software for an answer to delegating work to it.
For consumers, the appeal is easy to understand. Why visit several websites, navigate unfamiliar interfaces, and respond to repeated prompts if an agent can take care of the task?
For the companies behind those websites, the implications are much less comfortable.
I see a risk that personal agents become a new generation of adblockers. An agent instructed to minimize data sharing could systematically decline optional tracking. It could also complete a task without exposing its user to the pages, advertising, and brand experiences that traditionally supported the transaction.
This does not mean all data collection disappears. A purchase still requires information to fulfil it. But many of the signals businesses rely on for measurement, personalization, and understanding their customers could become much harder to obtain.
The consequences extend beyond advertising revenue. If brands see less of the customer journey and have fewer opportunities to establish a direct relationship, they may become even more dependent on intermediaries for discovery and acquisition.
Many businesses already rely heavily on Meta and Google to reach their customers. What happens if the next generation of platforms also becomes the primary place where customers express their needs, compare options, and decide whom to buy from?
My concern is that knowledge of the customer, and influence over the relationship, could become even more concentrated.
There is an important distinction here: people should absolutely be able to refuse tracking and delegate that choice to an agent. The challenge is to make sure that delegation genuinely reflects their intentions, while allowing a transparent exchange with the businesses they choose to engage with.
Can an agent communicate a person’s preferences accurately? Can a company explain what information it needs and for which purpose? Can the resulting permissions remain specific, understandable, and revocable?
These are the questions we need to solve for agent-to-agent privacy interactions. Otherwise, we risk replacing today’s imperfect consent experience with decisions that are even harder for people and businesses to understand.
Building for what comes next
I remain excited about what this transition can make possible: less manual work for our clients, more useful services for consumers, and privacy choices that follow people across their interactions.
Getting there will require us to rethink our own product as seriously as we ask our clients to rethink their data practices.
At Didomi, our direction is to make privacy signals reliable and actionable wherever customer data is used, including when the system using it is an AI agent.
That feels like a natural continuation of the work we have been doing for years, and a substantial new challenge.
I’m eager to hear how you are approaching it. As agents take on more of the interaction, what will help your organization preserve trust and a direct relationship with its customers?
