Perspectives

From the Party Line to the Agent

Whose number is it, and whose agent? What a 1992 scenario study says about agents in 2026

From the Party Line to the Agent. Whose number is it? Wendy MacKinnon. The agent belongs to the person.

When I was growing up, our family had a party line. One telephone line was shared by several households. You listened for your ring, and when you picked up, a neighbour might already be on it.

The number belonged to the house, and sometimes to several houses at once.

Then
The house
1992
The car
2003 on
The person
1992

The number belonged to the car

By 1992 the number had started to move. While at Gemini Consulting, I worked on future industry scenarios for the wireless business of one of the Baby Bells. I found the material this week.

Cellular reached about 3% of the population, mostly for business. Most phones were still car phones, and the number belonged to the car. In an April 1992 working session, the company’s own experts put a short list of events more than ten years out. First on the list:

April 1992 working session

One-number-per-person is introduced.

Requires increased network intelligence.

A 1992 slide titled Respondents Believed Certain Events Would Occur Only Beyond a Ten-Year Time Horizon, with One-number-per-person is introduced highlighted.
The April 1992 slide, with the client’s name removed.

It took eleven years. Wireless number portability came to the US in 2003 and to Canada in 2007.

The method

The method, not the forecast

The forecast is the easy part to admire. The method behind it is the part worth reusing.

We worked from Peter Schwartz’s The Art of the Long View. It doesn’t ask you to predict the future. It asks you to build a few plausible futures, then choose signposts that tell you which one is arriving. In 1992 we called them milestones, and the closing recommendation was:

1992 next steps

Develop a process to monitor ongoing events.

That humility was earned. The word “internet” never appears in the 1992 material. The Mosaic web browser was still a year away. And the industry underestimated itself: in 1991, carriers hoped to reach 15% of Americans by 2000. They reached about 34%.

The forecast matters less than the method behind it.

Two things the 1992 work got right still apply.

People want the outcome, not the plumbing
Customers would care about the benefit, “such as voice access anywhere,” and be “indifferent to the technology used.”
Non-traditional players arrive early
“Early emergence of non-traditional players will dramatically affect” the industry. In 1992 that meant Motorola, computer makers and cable companies. In 2026 one of them is Meta, with an agent inside WhatsApp.
2026

The 2026 signposts

If I ran the same exercise on agents today, these are the signposts from the last twelve months I would mark:

01
Models that work for a day
Independent testing found frontier models completing tasks that take a human expert a full workday, about half the time. A year earlier it was under an hour.
02
Agents inside the apps people already use
Meta launched Muse in WhatsApp in the US in September. Instinct runs in iMessage.
03
Protocols moving to standards bodies
Agent payment protocols from Google, Mastercard and Coinbase moved into standards bodies this year.
04
Canada’s open banking framework
The Consumer-Driven Banking Act passed in March. Draft regulations followed in June, and the Bank of Canada is now consulting on accreditation.
05
Instant, irrevocable payments
Canada’s Real-Time Rail goes live on November 9.
06
Portability beyond banking
The federal budget proposes data portability for telecom and transportation.
The 1992 milestones slide above a signpost of 2026 events, topped by the question: Who does the agent answer to?
1992 milestones and 2026 signposts.

Each is a signpost on its own. Together they point to a world where an agent can see your data, make a decision, and move your money in seconds, with no way to pull it back.

Three scenarios

Who does the agent answer to?

Schwartz would ask which uncertainty separates the possible futures. For agents, I think it is this one. Three answers give three scenarios.

01
The platform’s agent
It belongs to the platform that built it. It is convenient and capable, and it serves the platform’s commercial interests first. The customer gets the outcome, but not the reasons.
02
The merchant’s agent
Sellers deploy agents that negotiate, recommend and close. Customers face a well-equipped agent on the other side with no equivalent of their own.
03
The customer’s agent
It answers to the customer, works within a mandate the customer sets, checks who it is paying before money moves, and keeps a record the customer can rely on.

These aren’t predictions, and parts of all three will exist at once. The question is which one becomes the default.

What to watch

The signposts to watch

A few events would tell us which way it is going:

The number belonged to the house, then to the car, then to the person. On a party line, you never quite knew who else was listening.

The next step is making sure the agent belongs to the person too, and answers to no one else on the line.

That is the scenario Twinvocate is built for: an agent that answers to the customer, with the mandate, the counterparty check and the receipt in code. Twinvocate is at the prototype stage, and no live payments or real customer data run through it.

I would welcome hearing from others who have done scenario work, in telecom, payments or anywhere else, about the signposts they are watching for agents.

Sources

Wendy MacKinnon is the founder and CEO of Twinvocate, a Toronto company building a customer-loyal AI agent and a certified-counterparty credentialing layer for agentic commerce.

All insights