AI 日常|科技工具|海外生活|成长记录
分享真实体验,也记录慢慢变好的过程
AI daily life · Tech tools · Life abroad · Growth notes
Sharing honest experiences and the process of becoming better
Hey @ycombinator — I’m applying to YC Startup School and wanted to submit Paxel as the optional evidence. I have a lot of coding-agent history, but the free quota hit daily_cost_exceeded before the full report finished. Could applicants get a slightly larger quota or ? 🙏
Time to say goodbye to overpriced phone bills. I was paying a whopping $200 every month for 4 lines on AT&T. Just saw that Google Fi is offering 50% off all unlimited plans for a full year for new accounts.
Their best plan, Unlimited Premium (which is perfect for international travel just in time for summer), drops from $160 to just $80/month for 4 lines. That’s only $20 per line! Even after estimated taxes and fees, it shouldn't go past $100. Absolutely incredible value. But only last for a year which is a bummer
I had the same question, especially for real coding work.
So I built a local Claude Code version of the Fusion idea: Fusion Deck.
It runs Opus + GPT via Codex + Gemini as a blind panel, then has Opus judge agreement, disagreement, and blind spots.
The extra part I wanted to test is workflow:
fuzzy ask → plan → repo context → panel judgment → verified steps → handoff
So far the interesting bit isn’t just “is the final answer smarter?”
It’s whether the panel + workflow reduces confident wrong turns in agent work.
Repo if useful:
https://t.co/6UyiE0lZNU
So the loop becomes:
fuzzy idea
→ plan
→ right context
→ panel judgment when needed
→ verified execution
→ handoff
Fable gave us the taste.
Fusion Deck gives it memory, context, and workflow.
Repo:
https://t.co/3b4Po9AQql
This was the spark.
We are getting Fable back — but with tools.
I built Fusion Deck: the OpenRouter Fusion idea running inside Claude Code, using the CLIs already on your machine.
A blind model panel for judgment.
A workflow layer for real agent work.
/fusion-plan turns fuzzy asks into real plans.
/fusion-context packs the right repo context instead of dumping the whole codebase.
/fusion-orchestrate splits big work into verified agent steps.
/fusion-review reviews code with the panel.
/fusion-handoff keeps continuity across sessions.
That brings back the Fable feeling:
not just autocomplete,
but judgment.
But I wanted more than a better answer machine.
So I added what I always wished Fable had for real coding work:
planning, context, orchestration, review, and handoff.
Fusion Deck runs a blind panel with:
Opus 4.8,
GPT-5.5 via Codex,
Gemini 3.1 Pro.
They answer the same question in parallel, blind to each other.
Then Opus judges:
where they agree,
where they split,
what they missed,
and what the final call should be.
https://t.co/FnKgm8FoPm
Switched to @ATT with a promo promising $100 Visa gift cards per line. Been over 2 years! Never received a single one. Followed up for months — CS said they "expired" and literally hung up on me. Filed a formal @BBB_US complaint for $500. Unacceptable. #ATT#ConsumerRights
转网到 @ATT,销售承诺每条线送 $100 Visa 礼品卡,结果一张都没收到。追问了好几个月,客服说"已过期不补发",然后直接挂了我电话。$500 的承诺就这么没了?已向 BBB 正式投诉,绝不认了!#ATT #消费者权益
btw the entire process — finding the right BBB listing, filling out the complaint, submitting — was handled automatically by @Perplexity's AI browser Comet. It even tried to log into my X to post this lol. $20/month, personally feel smarter than Manus (which I got free from Lenny Bundle). This is the future. #Comet #AI
顺便说一下,这整个 BBB 投诉流程全部是 @Perplexity 的 AI 浏览器 Comet 帮我自动完成的,它甚至想帮我登录 X 来发这条推文哈哈。$20/月会员就能用,比 Lenny Bundle 送的 Manus 好用多了,这就是未来!#Comet #AI