Background auto

The future of software is personal

4 min read

Imagine hiring a nutritionist whose approach you like. Their service prepares your meal plan and shopping list around your preferences, budget and week. I want to reach that service through a personal agent that already knows me. It keeps our history and uses the email, calendar and messages I choose to connect.

A travel confirmation arrives for Tuesday to Thursday. My agent notices the overlap with my meal plan and sends a question: “Does the hotel include breakfast?” I answer by voice: “Yes, and I'll eat at home Thursday.”

While I get on with my day, it checks the hotel's menu in the browser and asks the nutritionist's service to revise the plan using their agreed alternatives. It updates the plan and shopping list in the apps I use, then lets me know they're ready. The breakfast detail stays with this trip; my regular preferences carry into future weeks.

Trip confirmed

Tuesday–Thursday away

Your voice reply

Breakfast is included. Home Thursday.

Your agent Notification

Your nutritionist's plan is revised for the trip.

Meal plan

Tue–Wed
Breakfast at the hotel
Thursday
Chickpea curry at home

Shopping list

  • Chickpeas
  • Rice
  • Spinach

Tue–Wed home dinners removed.

Your agentFollowing Monday

You're home this week. Your usual breakfasts are back in the plan.

Illustrative service. Choices come from the nutritionist's agreed alternatives.

I pay for that person's judgment applied to my life, including the revisions between appointments. I can ask why a meal changed and learn from the decision, or get on with my week.

Buying the work

In Brazil, many people already sell their expertise through courses. I expect more of them to sell a service that applies that expertise for the customer. The nutritionist I follow could offer ongoing meal planning to people they don't currently have time to see individually.

I expect the personal agent to sit above the services I use, carrying my context between them. It could help a nutritionist plan my week, an accountant close my books or a legal service finish an assignment. Some jobs run between conversations. Others start when a message, a calendar change or what I'm doing makes them relevant. Where an app has no suitable connector, the agent could work through a browser or computer.

A person helping me would remember Thursday's change when we spoke again. I would expect my agent to do the same.

Serving more people

An expert may have more people asking for help than they can serve. I expect software to make some of that service affordable for customers who currently get a course, an occasional consultation or no help at all.

My travel confirmation, usual preferences and agreed meal plan are different records. Their relationships tell the agent which choices belong to this trip and which should carry into next week. Retaining the work lets the service revise the plan as my circumstances change.

The nutritionist still has to trust the result. If checking every revised week means planning it again, the service has saved little of their time. The provider remains responsible for completing the work and correcting mistakes.

Harvey's October 15 release organizes legal work into tasks with inputs and designated reviewers. Rogo lets firms apply their workflows to research, models and presentations. Pilot already sells monthly bookkeeping, including plans where a bookkeeper handles and reviews the accounts. A founder could sell tools to practitioners or operate a service that uses them.

Lovable, Codex and Claude Code make it easier to build the tools behind those services. The founder still has to find customers, gather their information, finish the work and handle revisions.

Choosing the service

My personal agent could also be where I choose and pay for these services. If I ask for help with meals, it could recommend a nutritionist whose approach fits what I need. Once I choose, I would expect it to keep using that service until I decide otherwise.

The company running the agent would influence who gets those introductions. It could put an unknown specialist in front of a customer or repeatedly recommend the same few. An expert who brings their own clients would still depend on continued access. I would want those clients to be able to take their ongoing plans to another assistant.

Where I would start

A provider could start through the messages and documents their customers already use, before an assistant such as Muse or Instinct becomes a place to buy the service.

I'd work with one expert and a few customers who already need the same kind of work. Sell a defined assignment, deliver it and record where the expert intervenes. Count the time spent chasing information and correcting mistakes, along with the time preparing the result.

With the nutritionist, I'd follow one client through two changed weeks. I'd want to see whether the revised plans arrived in time, whether they took less work to deliver and whether the client paid for another month. The missed context, repeated questions and corrections would tell me what to build next.

:sources 8
  1. Lovable: Partner program for experts and service providers, September 9, 2026
  2. OpenAI: Codex use beyond software engineering, June 2, 2026
  3. Anthropic: Agentic coding and domain expertise, June 16, 2026
  4. Harvey: Legal Work Orchestration, October 15, 2026
  5. Rogo: Custom agents, connectors and memory, May 2026 product update
  6. Pilot: Bookkeeping services and expert review, checked October 18, 2026
  7. Instinct: Personal assistant
  8. Meta: Introducing Muse
:commands

j/k scroll · gg top · G bottom

[/] previous / next post · b blog · H home

t cycle theme · ? commands · Esc clear and close