
“Is the bot development session stuck in a review loop? When can we go live?”
I asked GPT that question in November. Fix something, review it, and check it again. I had kept working through the necessary tasks, but I wanted a clearer point at which I could move forward.
I began prototyping in February, was running several bots by May, and continued revising and testing afterward. This final part of the series describes my thinking and the project’s status as of November 16, 2026.
The market keeps moving while I wait to finish
In October, I remarked that “by the time it is finished, the edge will be gone.” By edge, I mean an advantage that can help a trading approach make money.
However carefully I build the software, the idea I want to test may not remain useful indefinitely. Development and the market move on separate timelines.
Of course, launching sooner does not guarantee a profit. But if preparation continues indefinitely, I also gain no new information from live operation.
I wanted to work out how to move between those two realities. What needed checking first, and what could I learn while running it? Treating every task as equally important made it harder to decide when to proceed.
Restating the purpose
In an October conversation, I described the bot as support for the periods when I take a break from discretionary trading. I wanted it to trade while I stepped away from watching the markets for a few days to a week.
Returning to that purpose helped clarify the priorities. How much should it do while I am away? Can I understand the situation if something happens? How do I define the responsibility I am comfortable delegating?
There is no natural end to the list of features I could add. AI can suggest more improvements and checks. That is useful, but choosing what serves the current goal still falls to me.
I needed to define the conditions for completing a stage and running the software, alongside the reasons for continuing development.
Start small, keep the essential safeguards, and improve
In November, I explained how I wanted to proceed. The main idea was this:
I want to move into live operation on a small scale with the necessary safeguards in place, then improve the bot as I run it.
That expressed my priorities. I wanted to begin within a small scope, observe the results, and improve from there.
Doing that requires concrete pre-launch checks. If the conditions for proceeding remain vague, every review can expand the work again.
Separating issues that must be fixed before proceeding from improvements that can be made during operation became part of the development work itself.
Passing tests and being ready to operate
The recent work still includes implementation changes, tests, and preparing material for review. These are steps toward checking the intended behavior and moving to the next stage.
A passing test establishes something about the scope it covers. It does not, by itself, establish that the system will perform reliably over a long period in a live market.
Losing that distinction creates a mismatch between the expectation that “everything passed, so it must be finished” and the reality that some questions remain.
I am building on what has been verified and working toward operation under limited conditions. At the time of writing, I cannot report that this new stage of live operation has been completed or that long-term stable operation has been achieved.
Building with AI is still interesting
The long development process has not removed the interest I felt at the start. I can put an idea into words, see it become code, and observe how it behaves. That offers a different kind of satisfaction from discretionary trading.
Even as the range of tasks I can delegate to AI grows, deciding what is sufficient remains difficult. Safety, development speed, and information from actual operation all matter. I still have to decide which takes priority at a given point.
These months have also been about learning how to develop with AI. Asking for a specific fix and deciding how the entire project should progress are two responsibilities I have needed to consider separately.
What I want to write about next
Looking back from February, there have been several stages: creating the prototype, running multiple bots, addressing operational issues, and now defining the scope of verification so I can move forward.
It has not been a straight line to completion, nor a matter of endlessly rebuilding the same thing. Running the software raised new questions, and I made changes in response.
Next, I want to write about how much responsibility I could actually hand over. Did it reduce the time I spent watching the markets? What needed adjusting during operation? Did it feel worth continuing?
Those answers will come from operating it. For now, the plan is to put a boundary around the necessary checks, start small, and improve from there.
Continue the series
カテゴリー:Bot Development

1件の返信 »