A couple of years ago I could not have told you what an API was. Today a bot I built runs on a small cloud server every evening, checks the market against rules I wrote, and flags what's worth a look. I still don't have a coding background. What I have is a workflow.
Plain English in, working code out
I never sat down to "learn to code" first. I described what I wanted in plain English, got code back from an AI, ran it, watched it break, and pasted the error back in. Repeat until it works. Then make it explain every line until I understand why it works.
The first real version looked roughly like this:
# What I asked for, in plain English:
# "Every evening, check my watchlist. If a stock fits my rules,
# flag it. Never buy anything without asking me first."
def evening_run(watchlist):
for symbol in watchlist:
bars = get_daily_bars(symbol)
if fits_my_rules(bars):
flag_for_review(symbol) # a human makes the final call
Nothing clever. Watch, decide with rules, act. That's the whole shape of a trading bot. Everything else is detail on top of those three moves.
What the AI didn't do for me
It wrote code fast. It did not tell me what to build, which assumptions were wrong, or when "it runs" wasn't the same as "it's right." That part was on me, and it's where most of the real lessons came from.
What I'd tell another beginner
- Build the guardrails before the exciting parts.
- Make the AI explain its code. If you can't read it, you can't debug it.
- Keep the stack lean: a broker with an API, a small server, Python.
- Start before you feel ready. The mental block is harder than the building.
I wrote the full story, including the fears that almost stopped me and the mistakes I'd skip, in How I Built an Automated Trading Bot With No Coding Background. If you want the actual stack, the trading bot tools I recommend page has it.
This is education, not financial advice.
Top comments (1)
For an evening scanner, add a check that the last daily candle has actually closed in the exchange timezone. A second run on the same saved bars should produce the same flags. If bars are missing or stale, log a skipped symbol instead of treating that as a failed trading rule. Those checks fit your human-review design without adding order execution.