The Plan Can Change

I wanted to make a large claim small enough to test.

I expect more of us to give software an outcome and let it work out the steps. It will choose tools, change plans, and hand work to other systems. A small team could attempt things that once required a large organization. I want that future. I also want the limits we give software to hold as it finds ways to get things done.

I started with payments because they turn that ambition into a concrete decision. Consider a payout that times out. The original attempt may have paid, or it may still pay. Another provider is available. The software has a reasonable objective: get someone their money. Sending again might look like progress. It could also pay the same obligation twice.

A new route is not a new right to pay. That became the starting point for HALTSEAL. Idempotency protects retries within a provider’s defined scope. My question is what happens when the same obligation can be paid through a different provider, another account, or a manual fallback. Each path can work as designed while the combined result is wrong.

Conversations with payment operators and a bank risk leader have made me more careful about that question. Existing controls deserve to be understood. I want to start with a recent payout, trace every path that could still move the money, and find the decision that governs them together. If the existing system already does this well, I need to understand how.

In one technical conversation, a team described a system where software prepares a payment and a person retains the final signature. That helped sharpen the boundary. Preparing, approving, and executing are different moments, and a successful request does not necessarily establish what happened to the money. HALTSEAL’s decision has to be enforced where the action can still be stopped.

Building the core system forced me to make that precise. While the first attempt remains unresolved, the capacity committed to it remains spoken for. Switching providers does not replenish it. Restoring that capacity requires authoritative evidence that the original attempt neither paid nor can still pay. Even then, the replacement needs a fresh check against current authority. Closing one attempt does not approve the next.

HOLD carries a responsibility too. Someone may be waiting for money they earned. An operator may have to investigate, and a business may have promised payment by a certain day. Running Gooroo taught me to take those promises seriously. A useful control should explain why it is holding, what evidence is missing, and what would allow the decision to change. I intend to measure the delay and work it creates as carefully as the excess payments it prevents.

My Christian faith makes this personal. Love of neighbor includes people who never chose the software but still have to live with what it does. It includes the supplier waiting for payment and the operator trying to put things right. Their time and obligations belong in the design from the beginning.

The public demo and synthetic tests let people inspect and challenge the behavior. The next test is a real workflow with competent existing controls. It needs to show that the relevant paths can be governed, uncertainty can be resolved, and recovery works at an operating cost that makes sense. If it cannot, I want to know. That is how I intend to earn a wider scope: prove one boundary, then another.

I expect increasingly capable software to get better at finding another way forward. A failed provider may become a reason to switch routes. An obstacle may become a reason to involve another agent. That resourcefulness is part of the promise. It also means a limit attached to one tool or credential may be left behind. The limits need to follow the work.

The plan can change. Permission should not quietly expand with it. And when a person withdraws that permission, the system needs to honor the change while it can still prevent the action. I want people to entrust useful work to machines without spending the time they saved watching every move. Dependable delegation would give us room to attempt more, and room to live beyond the work.

I started with payments because I want the larger promise to be earned.