Pinned Tweet
Day 7
Had some technical issues getting trades to fire properly, so our automation missed a couple trades including a 3.7R win 😫.
All good though, here are a couple of wins from Thursday's Asia session.
The bot is up $800+ on a funded and $2K+ on an eval. Not bad for 1 week.
We've also been adding more strategies to the portfolio, currently sitting at 7 strategies, but only 5 are live right now. Our quant factory is doing WORK.
Coming along nicely.
3
328
On Friday we were ready to spend thousands of dollars hiring a user testing agency to test our app.
Now it's Wednesday and we're 50% done just using Claude Code. Definitely not hiring that agency now.
If you vibecode anything, here's how how to speed up your testing by 100X 👇
1
214
Step 3
Hook up Codex or Claude Code or even OpenClaw (whatever tf you use to code) to your MCP.
Congratulations, your coding agent can now navigate your app like any other user and can use that feedback loop to test (and fix) everything on your app.
1
95
Day 9
If I took this trade manually I 100% would have lost.
Automated trading with a quant strategy trained at the 1s level is wild.
The bot took a trim at 1R and exited the rest at BE.
1
2
152
Day 8
Today our bot passed 1 eval and made $750 on a funded.
To-date:
+$3,000 on eval (now new funded)
+$1,591 on funded
It took a (risked-down) trade in NQ London while we were sleeping for +6R and a trim + BE this morning on ES for +1R.
Pretty great start to the automation, can't wait to scale this up.
1
1
125
Over the last 10 years, our 8-strategy portfolio has done:
- 5719 trades (572/yr)
- 1309R (131R/yr)
- 48.21% Win Rate
This is what we're running our accounts right now.
1
3
119
😂😂 real coders aren't using sonnet or 5.2. They're using Opus or Codex 5.3. Not a good look when you have to exclude the best models to make yourself look good.
1
3
203
Keyboard-first is a forcing function for good UX. If you can't navigate it without a mouse, your IA has hidden complexity. Power users who live in shortcuts will find it fast anyway, better to design for it upfront.
Day 55 of building my app:
Improved keyboard accessibility of the desktop version. The goal is to make it fully accessible and usable without a mouse
#buildinpublic
1
132
The biggest MCP server mistake is cramming 20 tools into one server. Claude picks the wrong one half the time. Split by domain, name them obviously, and suddenly tool selection just works.
The 'too fast' framing assumes speed is the variable that matters. Job displacement, model opacity, and power concentration don't get solved by slowing the clock. They get solved by governance. Pace is a proxy for the real concern: who controls this and who's accountable.
TIME’s new cover: In a deeply divided nation, a new coalition is forming around one belief: AI is moving too fast. Inside the stories of nine Americans, across ideologies and professions, determined to slow down the technology reshaping daily life time.com/7377579/ai-data-cen…
1
86
The real unlock is that MCP gives Claude your actual schema context. It stops hallucinating field names and just writes queries that hit your real data. That's what used to take weeks, not the dashboard building itself.
I created 9 new dashboards in PostHog in 10 minutes using Claude Code + PostHog MCP.
Holy sh*t, this would have taken weeks back in the days.
112
All the OpenClaw malware drama and I'm just sitting here writing my own skill files locally. It's a markdown file with instructions, you can read every line. Idk why people pull random skills from a marketplace when you can write one in 10 minutes.
90
Every model is getting scary good at the first render. The part nobody has cracked yet is iterative editing without breaking everything else on the page.
gemini pro 3.1 ui gen is cracked
just one-shot a full landing page from
"professional saas landingpage for an ai agent provider. 8sections. use micro animations"
1
3
126
Day 1
My brother and I have been working on trading automation.
Yesterday was our first day running it on a funded account.
It took 2 trades, both didn't get to full TP, but they did get to TP 1. (First trade was 3 ticks from full TP 😫).
Anyway +$600 on Day 1, let's see how far we can take this.
2
189
Been defaulting to Opus for everything but if Sonnet handles conditional routing better with fewer tokens, that changes the math real fast
Sonnet 4.6 is here. Less than two weeks after Opus 4.6.
Does it live up to the hype?
Yes it does.
We ran @Claude Sonnet 4.6 head-to-head with 4.5 and Opus 4.6 on our Workflow Automation Benchmark.
On calendar-CRM coordination and conditional scheduling routing, Sonnet 4.6 was the only one that executed correctly end to end. The kind of task where the model has to evaluate conditions, pull the right records, and pick the right path…even a 10% error rate could lose a deal.
Overall, Sonnet 4.6 is a clear step up from 4.5, uses fewer tokens, and lands close to Opus performance in our testing.
So when would I prefer which?
I'd start with Sonnet 4.6 for workflow automation: conditional routing, calendar-CRM coordination, document and contract flows. Across our benchmarks the reasoning was close enough to Opus on those tasks, with far fewer tokens. That's my default for most Zaps.
I'd reach for Opus 4.6 when the workflow needs code generation or heavy coding support, when it spans a lot of tools and the model has to synthesize across several sources, or when steps can fail and you need the model to recover and try a different approach. Opus still has the edge there.
We shared these results with @AnthropicAI, and are excited to keep partnering on future launches so we can help more shared customers get the most out of Claude inside @Zapier.
Learn more here: zpr.io/pBHPReyxnnjM
1
139





















