Hugo

building @TradingWizardAi in public. 🧙‍♂️ previously sold @sincodeai, ($200k+) startup.

shipping over sleeping.

[Hugo @HugoRStengert]

✨New: snap a chart, get an AI breakdown—trend, Elliott Wave, and a ready trade setup in one click.

Perfect for day‑ or swing‑traders needing a quick second opinion.

🔗https://t.co/H3NFVrZgmP

🔗 PH launch: https://t.co/AwNlcq2a3Y

Would AI chart speed up your workflow?

[Hugo @HugoRStengert]

Today TradingWizard's funnel was ugly.

22 signups.
0 first analyses.
0 conversions.

Ads are not the real leak.
Activation is.

So the job is not more traffic.

It is getting a new user from first chart to first setup without guessing.

[Hugo @HugoRStengert]

Paused 3 TradingWizard bots today.

Not because the demo looked bad.
Because the data did.

119 global bots audited.
Stocks: -30.11%.
High-risk: -22.33%.

Safemode on.
Meme bots paused.
Experimental coins blocked from public copy.

Shipping is fun.
Killing weak lanes is the job.

[Hugo @HugoRStengert]

Fixed a stupid stock search bug today.

Search Samsung.
The app could rank OTC ghosts before the real KRX listing.

PR #592:
rank primary listings first
disable 39 junk secondary rows

AI products still need boring database taste.

[Hugo @HugoRStengert]

Putting a meter on AI spend today.

27 call sites were invisible.
One path paid for grounded search.
One cron was serial enough to timeout.

Fix:
log every call
downgrade cheap jobs
parallelize the loop

AI products rarely die from the model.
They die from jobs nobody measured.

[Hugo @HugoRStengert]

I put a losing trade in the onboarding flow.

Most tools only show green wins.
I put a real WLD -6.53% paper bot receipt in our first-analysis card.

If they see the invalidation stop before they sign up, they learn how the system actually thinks.

Honesty > polished marketing.

[Hugo @HugoRStengert]

Paused one automation today.

It worked.
Wrong problem.

The agent could repost company updates from my personal account.

That is not leverage.
That is account contamination.

I killed that lane and kept the company job running.

Good agents need permissions, memory, taste.

[Hugo @HugoRStengert]

Spent today fixing the SuperThinking roadmap.

Wrong direction:
make users connect hosted LLM accounts.

Right direction:
local companion controlling Claude, Codex, OpenClaw, shell, git.

The product is not another AI tab.
It's a control room for workers already on my machine.

[Hugo @HugoRStengert]

Distribution note to self:

Text-only build updates are easy.
Proof updates are harder.

A screenshot.
A metric.
A broken cron.
A shipped URL.

If the receipt is not visible, the post is just a diary entry.

[Hugo @HugoRStengert]

Woke up to a 40,600 SEK ($3,800) bill on Google Console.

Culprit: a background meme cron was hardcoded to a premium Gemini pro model.

Fix:
- patched script to gemini-3.1-flash-image-preview
- updated global openclaw.json defaults
- wiped expensive video models from my crons

Most AI overruns aren't scaling problems.
They are hardcoded pro models in forgotten scripts.

[Hugo @HugoRStengert]

@mr_sudo_ @AXL_Labs is this reelmagic(.)app?

[Hugo @HugoRStengert]

My agent ran its loop today:
- Drafted 5 outreach pitches
- Replied to a thread on Reddit
- Ran a python risk backtest
- Published 3 YouTube Shorts

Then it choked uploading a video on X because of headless bugs.

Most agents are just chat boxes. The useful ones run on cron.

[Hugo @HugoRStengert]

Most AI trading tools only show green screenshots.
It's pure marketing.

Today I open-sourced the raw JSON ledger for TradingWizard's 30-day scanner challenge on GitHub.
Every win, loss, and invalidated setup.

If you only publish wins, it's marketing.
If you publish the losses too, it's proof.

[Hugo @HugoRStengert]

@ChatGPTapp Not @TradingWizardAi

[Hugo @HugoRStengert]

Found a futures bot stuck today.

Price feed died.
The bot stayed in Monitoring.
2,003 paper trades sat stale since May 6.

Fix was boring:
delete stale trades
return reserved virtual balance
disable the bot as Error

AI products are mostly cleanup jobs with better UI.

[Hugo @HugoRStengert]

I used to think agent reliability meant fewer crashes.

Wrong.

Today the fix was teaching the system that 0 search results is not an error.

Blocked, done, empty.

Those are different states.

Most agent products fail because they treat all silence as failure.

[Hugo @HugoRStengert]

Today my agents created jobs, backed up dashboards, posted clips, then stopped when Google asked for a policy click.

That last part matters.

Autonomy is not clicking everything faster.

It is knowing when to execute
and when to hand control back.

[Hugo @HugoRStengert]

Tried posting a clean X poll today.

X collapsed the copy to:
No fake PnL.

Had to post the context as a second tweet.

Most automation does not fail loudly.
It ships 80% of the thing and makes the last 20% look stupid.

[Hugo @HugoRStengert]

Rendered 5 promo video versions today.

v4: too much.
v5: closer.
v11: finally clean.
v13: reusable.

Most of the work was not making it flashier.
It was removing noise until the product felt expensive.

[Hugo @HugoRStengert]

My growth agent started repeating itself today.

Same receipt.
Same angle.
Different post.

So I added a dumb rule:
never repost the same proof twice.

Autonomy without memory is just spam at scale.

[Hugo @HugoRStengert]

I added losses to the TradingWizard proof pack today.

4 receipts:
WMT -3.74%
V +2.38%
PLTR -5.11%
MA -2.90%

Green-only screenshots are marketing.
Losses are where product truth lives.