Your Customers Are Getting AI Assistants. You Should Too.

If you run a business, sit on a board, or need to understand your customers, get a personal AI assistant and use it to help run your life. You cannot delegate this assignment. You need to experience how these tools change behavior, where they fall short, and what new businesses their limitations will create. You also need to understand how much surveillance people may willingly accept in exchange for the convenience. Let’s explore.

What Has Changed

Previous generations of digital assistants were useful for setting timers, playing music, checking the weather, and performing a limited set of transactions. Getting them to do something more complicated often required more effort than doing it yourself.

The current generation can maintain context, work across connected services, and carry an assignment through multiple steps. Depending on the product and the permissions you grant, an assistant can read relevant correspondence, research alternatives, prepare work, and help execute a plan.

Persistence matters. You don’t have to start every interaction by explaining who you are and what you’re trying to accomplish.

My assistant, Hal, and I are already handling work together that would previously have required a human assistant. Context from earlier conversations helps with later assignments. Corrections make the next interaction more useful. Work can continue beyond the conversation that started it.

There are still plenty of things that require human judgment. Once you experience this kind of help, your expectations change. You begin looking at ordinary software and wondering why you still have to do so much of the work. Your customers will start asking the same question.

Pick Your Assistant

Consumer-friendly:

• Meta’s Muse combines ongoing context with browser-based work and connected services.
• OpenAI’s dots are persistent assistants with cloud computers and connections to other apps.
• Grok Bot from SpaceXAI (formerly xAI) offers hosted agents that share a cloud computer for each user.
• Instinct by Spear Street Technology emphasizes interaction through text and calls, with phone and computer capabilities.
• Gemini Spark (renamed Task mode) from Google can work on tasks in the background and connect to services like Gmail, Calendar, and Drive.
• Claude Duet from Anthropic is a proactive personal assistant that can monitor connected apps, prepare briefings, and work with you through a message-based conversation. At this writing, it is only available through a limited research preview.

For the more technically inclined:

• OpenClaw provides a self-hosted, model-agnostic approach. You take on more responsibility for how it runs and what it can access.
• NVIDIA’s NemoClaw belongs in a different category: It is early-preview infrastructure for running agents with security controls, rather than a comparable consumer assistant you simply sign up for.

These products have different permissions, interfaces, and levels of maturity. This is a starting list, not a market-share ranking or a recommendation to connect your entire life to whichever one has the best demo.

Choose something you can actually use. A technical setup you never finish will teach you considerably less than an accessible assistant you work with every day.

Experience the Consumer’s Side

Imagine asking your assistant to replace an appliance. It knows the space available, your budget, the features you care about, and the problems you had with the last one. It can help compare suitable products, check delivery conditions, and bring you a recommendation to approve.

Now imagine that you sell appliances. How much of your carefully designed customer journey did that person experience? Which parts of your advertising reached them? What happened to the product you wanted them to consider but their assistant rejected? The balance of information is changing. Consumers gain the ability to compare more options and investigate more details without doing all the work themselves. Some of the friction that protects incumbent businesses becomes easier to overcome.

That doesn’t mean every assistant will always choose the cheapest product. People have preferences that extend well beyond price. An assistant with useful context may know that you will pay more for a particular brand, a particular color, or delivery on a particular day. The commercial challenge becomes more interesting when an assistant can carry those preferences into a purchase decision. To understand that challenge, you need to be the customer. Watch what your assistant considers. Ask why it excluded something. Notice what you stop bothering to do yourself.

Your Measurement Tools Will Miss Part of the Journey

An assistant can research products in a cloud browser, return a shortlist in a conversation, and use an API to complete a purchase. The customer may never see your landing page. A page loaded by an agent does not establish that a person saw the ad on it.

Browser design complicates the measurement. An ephemeral session can discard the cookies that connect visits. A persistent agent browser can retain its own cookies and logins, while remaining separate from the customer’s usual browser. Some agents can also use an existing browser profile. You cannot assume one architecture across this market.

Mobile advertising identifiers, such as Apple’s IDFA and Android’s advertising ID, create another mismatch. A cloud browser does not automatically carry the customer’s phone identifier. Tracking permissions and platform restrictions still apply. A familiar device-based identity graph may tell you very little about the agent doing the shopping.

Agent identification is improving. Web Bot Auth uses cryptographic signatures to authenticate automated requests. That helps a site recognize who is connecting. It does not tell a marketer which ad influenced the customer, whether the customer saw a recommendation, or how much authority the agent has for this particular purchase.

Marketers will need ways to distinguish authorized shopping agents from scrapers and fraud, connect research to purchases with the customer’s permission, and avoid counting the agent and the person as two customers. They will need reporting on whether an offer entered a shortlist, which sponsored recommendations actually reached a person, and which sales were incremental. Consent, frequency limits, returns, attribution and independent auditing all need to work across agent, merchant and advertising systems.

Pieces of this infrastructure already exist. Consistent implementation across competing services remains a practical problem. An agent’s provider may see much more of the decision than the merchant sees.

This creates another opportunity for new businesses. It also creates a budgeting problem for yours. An order can reach your system while the process that produced it remains largely invisible to your marketing team. Ask your team to trace one purchase made with your assistant. Have them separate what they can observe from what their attribution model assumes. The gaps will make this discussion considerably more concrete.

Pay Attention When It Fails

The failures are part of the assignment. An assistant may remember an outdated preference, misunderstand an instruction, or confidently recommend something that doesn’t meet your needs. It may get stuck at a login, encounter an incompatible website, or finish most of a task and leave you with the difficult part. Remembering a preference also does not establish permission to act. “I like this hotel” and “book a nonrefundable room” are very different instructions. Keep track of where you intervene. Those interventions reveal both the current limits and the opportunities.

There are businesses to build around reliable authorization, verifiable product information, agent-accessible customer service, and completing transactions safely. There are opportunities in helping people inspect what their assistants remember, correct mistaken assumptions, and move useful context between services. Some problems will disappear as the underlying technology improves. Others involve identity, accountability, commercial relationships, or trust. Learning which is which requires getting your hands dirty.

The Surveillance Bargain

The most consequential part of this story is the information an assistant needs to become truly useful. That context can include your conversations, relationships, calendar, purchases, unfinished plans, and the reasons behind your decisions. An advertising system may infer that you are considering a vacation. Your assistant may know why you need one, who you want to travel with, what you can afford, and which compromises you are willing to make. The difference in depth is extraordinary.

Product architecture and privacy policies matter. Local processing, cloud processing, retention, advertising use, and access controls are different questions. Giving an assistant access to information does not automatically mean that information is sold to advertisers or available to the government. A richly connected assistant can create an extraordinarily detailed account of a person’s life. A breach, misuse, or compelled disclosure could expose far more than a history of clicks. In the depth of information we voluntarily assemble and entrust to a service, this could make today’s surveillance economy seem quaint. Why would people accept that bargain? Because the convenience can be extraordinary, too.

You need to feel that tradeoff personally. Notice the moment you consider granting another permission because the assistant could do something useful with it. Then ask what you have authorized, what you understand about the consequences, and whose information besides your own may be involved. Your customers will face these decisions. Many will make them quickly.

Do the Homework

Start with a bounded assignment and information you are entitled to share. Plan something. Research a purchase. Organize a personal project. Keep consequential actions subject to your approval. Respect your company’s security policies and other people’s privacy. Better yet, do this on your personal devices only. Navigating the personal vs. professional data issue will be one of the most instructive aspects of this assignment.

Importantly, using a personal device does not give you permission to share company information. Your work email, calendar, customer records, and confidential documents are governed by your employer’s policies, regardless of which device you use. The same is true of other people’s information. Before connecting an account or sharing a document, make sure you have the authority to do so and understand what the assistant’s provider can retain and use.

Stay With It

You’ll need to keep using your new AI assistant long enough for context to accumulate. Pay attention to what saves time, what creates work, what you begin to expect, and what access you become tempted to grant.

The particular assistant you choose matters less than completing the exercise. These products will change. The habit of delegating parts of everyday life is what you need to understand.

What happens when your customers become more capable, more demanding, and more willing to entrust intimate context to an intermediary? You will have a much better answer after you have experienced it for yourself.

Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.