A DataSapien point of view · For brands and partners
Brand‑Safe Agents
The path to making customer apps agentic, on device.
Start with low-risk journeys today. Add more agency as on-device AI matures. Personal context can stay on the phone throughout.
DataSapien is an on-device AI SDK and a no-code Orchestrator.
Agents are starting to sit between a brand and its customer
Large consumer assistants such as Muse, GPT and Grok act for the customer from the cloud. In September 2026, OpenAI launched Dots. These assistants search, compare and choose before the customer opens the brand’s own app.
The risk is that the brand loses its direct link to the customer. This happens in four steps.
The app goes quiet
The customer asks an assistant instead of opening the brand’s app.
The relationship moves to the assistant
The assistant holds the conversation, the history and the trust.
The brand becomes a supplier
Search sent customers back to the brand to buy. Agents complete the purchase themselves, so the brand becomes just an API the assistant calls.
Loyalty loses its meaning
When only the assistant sees the points, they become just a discount.
The same shift is also an opening. A brand with its own agent can serve its customers better than any outside assistant, inside its own app.
Where the agentic future could land
A few giant assistants control every relationship
Every customer relationship goes through a few cloud assistants. Brands compete for the assistant’s attention.
Assistants lead discovery.Prepared brands keep the relationship.
Customers ask an assistant to explore and compare. Brands with an intelligent app of their own keep the relationship after the customer chooses them.
At Booking.com, referrals from AI assistants are still “significantly below 1%” of room nights, and that share hasn’t moved in months, even though Booking appears often in AI answers.
Source: Booking Holdings Q2 2026 earnings call, 4 Aug 2026.
Intelligent brand apps
A new kind of brand app. The agent lives inside it, uses the customer’s context on the phone, and the relationship stays with the brand.
Cloud assistants become a threat when a brand has no agent of its own. DataSapien gives the brand a safe path to build one, step by step.
This matters most for brands with deep operational knowledge and customer trust: grocers, banks, airlines and health providers.
DataSapien helps brands move to the right: keeping a direct relationship with the customers, patients, members and citizens they serve.
Every agent needs a harness
By agent, we mean anything that acts on a customer’s behalf: simple software, an AI model, or a person.
-
1
Goal
What the customer wants done.
-
2
Information
Human and business context, combined on the phone (the Golden Context).
-
3
Decision
Choosing what to do.
-
4
Action
Doing it, with permission.
Domain knowledge · Orchestrator · Agentic experience design · SDK in the host app · Connected APIs · Brand rules
The big AI firms are racing to own the decision. DataSapien gives brands the harness: the tools to design, run and control an agent inside their own app. At its heart is information nobody else has: human and business context, combined on the phone.
The ladder below shows how much of the decision and the action a brand hands to its agent.
What “agentic” actually means
When Muse or Dots acts, the customer carries the risk. When a brand’s agent acts, the brand carries it.
DataSapien published the Agentic Ladder in July 2025, before “agentic” became a popular word. It sets out six levels, from manual to general.
The ladder measures how much a journey decides and acts for the customer. Risk rises with each step. Brand‑safe journeys run at A2 today, and A3 is next.
The right intelligence for each step
Each journey uses the least risky intelligence that gets the job done, matched to the customer’s device. That can be simple rules, the phone’s built-in AI, or the brand’s own small language model. Journeys move up the ladder as the tech matures. The brand sets the limits.
Software has always done A0 and A1. What’s new is the context it can now use.
Deterministic: same input, same output. Probabilistic: a trained model estimates. Generative: a model creates new content.
Builds a 7-day meal plan for the customer to choose from.
Calories burned (Apple Health), food likes (YouTube), the brand’s catalogue
Deterministic. Rules, catalogue search and matching.
With permission, adds the chosen plan to the customer’s calendar with reminders.
The customer’s calendar
Deterministic. Rules and API calls.
With approval, logs meals from photos and updates the day’s calorie balance.
Photos taken on the phone, activity (Apple Health)
Deterministic + probabilistic. On-device vision model on capable phones. On older phones, the brand’s cloud AI or a self-reported step.
Tracks live context, anticipates needs and adjusts the plan, within the limits the brand sets.
Live calorie burn and food intake during the consumption journey, calibrated with history held on the phone.
Generative. The phone’s built-in AI or the brand’s own small language model, with memory and rule guardrails.
Given a goal, such as a healthier month, plans meals and completes the weekly shop, once the brand judges the risk is worth it.
Deep life context, processed by on-device intelligence. Only abstracted tasks go to the cloud.
Generative. On-device intelligence works on deep life context. The brand’s own cloud AI handles abstracted tasks. All within rule guardrails.
Even simple journeys get better. A next best action that uses on-device context beats an advanced agent working from a central CDP.
What makes an agent brand‑safe
Bounded
It acts only within the journey the brand designs. Guardrails contain the risk of a model getting something wrong, and the platform is built with protections at every layer.
Auditable
Every journey, rule and change in the Orchestrator has permissions and an audit trail. What happens on the phone stays private to the customer.
On the device
It runs where the customer’s context already lives, so personal context never needs to move.
“Agency at design time, guardrails at runtime.”
The marketing team designs the journey. The device runs it within the limits the brand sets.
The market splits in two.
The intelligence goes to the data
Golden Context is built on the customer’s phone. The brand’s knowledge goes to the device. Nothing personal needs to leave it.
Human Context
Live signals on the customer’s device.
Business Context
The brand’s CDP and CRM.
Golden Context
Joined on the device, never pooled.
What brands can do today
Brands can now use context that could never be stored in one central place.
Used, not shared
Signals customers won’t share with a brand, but will let their own phone use for them.
Too sensitive to store centrally
Context no brand would want sitting in its own database.
Limited by terms
Data that app and platform terms keep off central servers. On the device, it can be used within those terms.
Limited by local law
Data that cannot cross a border. On the device, it never has to.
Live phone context
What the phone senses in the moment. The cloud never sees it.
Combined on the device
Data that can’t be mixed on a server. On the device, it can surface new insights.
- When it runs on the device
- No cloud AI cost for each action
- Works offline
- Cost stays flat as usage grows
How it fits the existing stack
DataSapien does not replace the CDP or CRM. It helps a brand get more from them, in the one place they could never reach: the customer’s device.
Business context and orchestration
The brand decides whether, when and what moves to the device.
The brand‑safe agent
The device builds Golden Context and runs the flow the brand designed.
What each stakeholder gets
For each person in the buying team. Find your role below.
Keep the relationship and act on live context, to earn trust and lifetime value.
Golden Context, without pooling personal data.
One SDK. Auditable journeys. AI only where it earns its place.
Control what data comes to the centre, and which personal data never leaves the device.
No cloud AI cost for on-device steps.
Flows built visually. No custom machine-learning pipeline.
What you bring to a client
Fits the existing stack
An on-device SDK and a no-code Orchestrator that sit alongside the client’s CDP, CRM, apps and cloud AI.
Ready for security and legal review
Admin control over what data reaches the centre, and an audit trail of every journey and change.
A method clients can follow
The Agentic Ladder gives a clear path from low-risk journeys today to more agency as the technology matures.
Gets better with every use
Each use deepens context on the phone and earns more trust, so returns grow for customers and brands.
See it for yourself
See a brand‑safe agent running on a real phone. Then start a pilot inside your own app.
The brand’s own agent, not a borrowed one.
Brand‑safe, on the device, using context the big cloud assistants never see. It keeps getting better, safely.
Humans at the centre. Not data centres. DataSapien.
