Not Caleb Dean

Developer. Before it was cool

[Not Caleb Dean @FakeCalebDean]

The original document, both pages. PDF: https://t.co/bR2ywzeSg6

[Not Caleb Dean @FakeCalebDean]

Well actually I lied I do have a degree. An MBA from Indiana Institute of Technology lol

[Not Caleb Dean @FakeCalebDean]

I started by talking to a chatbot.
Today I run a system of AI agents that researches, builds, applies, trades, guards my accounts, and verifies its own work — while I sleep.

No degree. No team. No funding. This is the full record. 🧵👇

📍 𝗣𝗛𝗔𝗦𝗘 𝟭 — 𝗧𝗵𝗲 𝘁𝘆𝗽𝗲𝘄𝗿𝗶𝘁𝗲𝗿 𝗽𝗵𝗮𝘀𝗲
Copy-paste prompts into ChatGPT. Felt like magic. It wasn't.
A chatbot that can't touch your files, your browser, or your inbox is a very smart typewriter.
The first real lesson: 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝗶𝘀𝗻'𝘁 𝘁𝗵𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁. 𝗧𝗵𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝘄𝗿𝗮𝗽𝗽𝗲𝗱 𝗮𝗿𝗼𝘂𝗻𝗱 𝗶𝘁 𝗶𝘀.

📍 𝗣𝗛𝗔𝗦𝗘 𝟮 — 𝗕𝗼𝘁𝘀 𝘁𝗵𝗮𝘁 𝗱𝗼 𝗼𝗻𝗲 𝘁𝗵𝗶𝗻𝗴, 𝗽𝗲𝗿𝗳𝗲𝗰𝘁𝗹𝘆
I stopped asking questions and started shipping tools:
• 𝗕𝗧𝗖 & 𝗘𝗧𝗛 𝟭𝟱-𝗺𝗶𝗻𝘂𝘁𝗲 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱𝘀 — live signal engines with their own audio alerts. When an A+ setup fires, my speakers tell me before my eyes do.
• 𝟯𝗟𝗦 𝗘𝘅𝗽𝗲𝗿𝘁 𝗔𝗱𝘃𝗶𝘀𝗼𝗿 — my own three-level strategy written in MQL5, running unattended on MetaTrader.
• 𝟰𝗛 𝟮𝟬𝟬-𝗠𝗔 𝗽𝘂𝗹𝗹𝗯𝗮𝗰𝗸 𝘀𝗰𝗮𝗻𝗻𝗲𝗿 — price above the 200, wick tagging it, RSI under 50, closed candles only. No repainting, no excuses.
• 𝗣𝗼𝗹𝘆𝗺𝗮𝗿𝗸𝗲𝘁 𝗰𝗼𝗽𝘆-𝗯𝗼𝘁 + 𝘄𝗮𝘃𝗲 𝘀𝗰𝗮𝗻𝗻𝗲𝗿 — mirrors sharp wallets and maps where the money is moving.
• 𝗢𝘃𝗲𝗿/𝗨𝗻𝗱𝗲𝗿 𝗦𝗵𝗮𝗿𝗸 — sports line scanner that hunts mispriced totals.
• 𝗘𝗗𝗚𝗔𝗥 𝘀𝗰𝗮𝗻𝗻𝗲𝗿 — reads SEC filings the second they drop.
• 𝗗𝗶𝘀𝗰𝗼𝗿𝗱 𝘀𝗰𝗮𝗻𝗻𝗲𝗿 — watches the servers I care about so I don't have to.
• 𝗦𝗔𝗠.𝗴𝗼𝘃 𝗯𝗶𝗱 𝗳𝗶𝗻𝗱𝗲𝗿 + 𝗯𝗶𝗱 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗼𝗿 — a Chrome extension that surfaces government contracts and drafts the response.
Every one taught the same thing: the AI is the easy part. The hard part is the data, the edge cases, and proving the bot isn't quietly lying to you.

📍 𝗣𝗛𝗔𝗦𝗘 𝟯 — 𝗔𝗴𝗲𝗻𝘁𝘀 𝘁𝗵𝗮𝘁 𝗮𝗰𝘁 𝗶𝗻 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹 𝘄𝗼𝗿𝗹𝗱
This is where it stopped being scripts and started being labor.
• 𝗝𝗼𝗯-𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗮𝗴𝗲𝗻𝘁𝘀 — read the posting, tailor the answers, drive the actual hiring site, submit, then open my Gmail and confirm the receipt exists. 14 applications in 40 minutes. Every one verified by the email, not by the bot's word. Separate lanes for FedEx, xAI, Memphis-metro roles, and fully-remote Greenhouse jobs.
• 𝗔𝗽𝗲𝘅 + 𝗕𝗼𝗼𝗸𝗺𝗮𝗽 𝗚𝘂𝗮𝗿𝗱 — a risk layer that watches my funded trading account and stops me from breaching the rules that would blow it.
• 𝗙𝗹𝗮𝘀𝗵-𝗨𝗦𝗗𝗧 𝗚𝘂𝗮𝗿𝗱 — detects fake "flash" stablecoin tokens before they touch a wallet.
• 𝗗𝗲𝗮𝗹 𝘀𝗰𝗼𝘂𝘁 — crawls business marketplaces, filters by real revenue and age, blacklists the junk, and hands me a shortlist of finalists to verify.
• 𝗗𝗼𝗿𝗺𝗮𝗻𝘁 𝗜𝗣𝘃𝟰 𝗹𝗲𝗮𝗱 𝗹𝗶𝘀𝘁 — found unused address blocks in the ARIN region and turned them into a broker lead list.
• 𝗙𝗼𝗿𝘄𝗮𝗿𝗱-𝗱𝗲𝗽𝗹𝗼𝘆𝗲𝗱 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝘁𝗿𝗮𝗰𝗸𝗲𝗿 — finds FDE roles hiding under other titles.
• 𝗧𝗿𝘂𝘁𝗵-𝗳𝗶𝗻𝗱𝗲𝗿 — pulls breaking news and cross-references it across sources before I believe any of it.
• 𝗬𝗼𝘂𝗧𝘂𝗯𝗲-𝗹𝗲𝗮𝗿𝗻 — reads a video's transcript, extracts the actual steps and claims, and files it in a library I can search months later.
• 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗳𝘂𝗻𝗻𝗲𝗹 𝗺𝗼𝗱𝗲𝗹𝗲𝗿 — finds a live offer, matches it to a traffic source, models the funnel by EPC, and returns a go/no-go with hard kill thresholds.

📍 𝗣𝗛𝗔𝗦𝗘 𝟰 — 𝗦𝗨𝗣𝗘𝗥𝗔𝗚𝗘𝗡𝗧
Then I built the thing that builds the things.
SUPERAGENT is my own orchestrator on top of Claude Code. It:
→ takes a goal, plans it, and spawns up to 16 subagents that work in parallel
→ drives a live Chrome session with my logins — clicks, fills, reads, verifies
→ reads my email, calendar, Drive, and brokerage account through real connectors
→ routes heavy reasoning to the strongest model and cheap scans to the cheapest one
→ runs every task as a closed loop: propose → validate → execute → feedback
Six hardening layers make it trustworthy:
1. Cheap check before expensive check — nothing runs the real test until it passes the fast one
2. Typed failures — every error is classified (malformed, rejected, crash, silent-fail) before a retry is allowed
3. A completion supervisor that climbs an escalation ladder instead of thrashing
4. A silent-fail watchdog — 𝗲𝘃𝗲𝗿𝘆 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗶𝘀 𝗿𝗲-𝘃𝗲𝗿𝗶𝗳𝗶𝗲𝗱 𝗯𝘆 𝗮 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗮𝗴𝗲𝗻𝘁 𝘁𝗵𝗮𝗻 𝘁𝗵𝗲 𝗼𝗻𝗲 𝘁𝗵𝗮𝘁 𝗯𝘂𝗶𝗹𝘁 𝗶𝘁
5. A live task board so I can watch the swarm work in real time
6. Instruction-source discipline — nothing it reads on the web can give it orders; only I can
It keeps a memory of 97 logged lessons from its own past runs. Every run starts smarter than the last.
I wrote the architecture up as a whitepaper: "𝗩𝗲𝗿𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻-𝗕𝗼𝘂𝗻𝗱𝗲𝗱 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆." It also lives on a USB drive that boots the same system on a Mac.

📍 𝗣𝗛𝗔𝗦𝗘 𝟱 — 𝗖𝗿𝗲𝗱𝗲𝗻𝘁𝗶𝗮𝗹𝘀
22 of 22 Claude Academy courses. Google Skills badges. Completed in days, not months — by the agents I built, supervised and audited by me. Every certificate verified in my inbox.

𝗪𝗵𝗮𝘁 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗰𝗵𝗮𝗻𝗴𝗲𝗱
2024: I asked AI questions.
2026: I give AI goals and audit the receipts.
The distance between those two sentences is the entire skill — and almost nobody has crossed it yet.

𝗪𝗵𝗲𝗿𝗲 𝘁𝗵𝗶𝘀 𝗴𝗼𝗲𝘀
Everything above was practice. Reps. Infrastructure.
The next thing I build is the one that gets monetized to the highest degree possible — a system that finds the opportunity, builds the product, sells it, and proves the money landed, without me in the room.
That's the difference I'm here to make. Not a tool. A machine that makes a living.

[Not Caleb Dean @FakeCalebDean]

The original document, both pages. PDF: https://t.co/bR2ywzeSg6

[Not Caleb Dean @FakeCalebDean]

The original document, both pages. PDF: https://t.co/bR2ywzeSg6

[Not Caleb Dean @FakeCalebDean]

The theory behind SUPERAGENT. An agent's self-report comes from the same process as its work, so the errors correlate. V1 builds: code is execution-checkable. V2 researches: claims must be corroborated. Two proof obligations, two architectures. https://t.co/FVGnDeGcg7

[Not Caleb Dean @FakeCalebDean]

As for the situational awareness portfolio...how did he get liquidated when he was supposedly up a huge percentage and caught infrastructure plays before the big run up...something doesn't make sense here

[Not Caleb Dean @FakeCalebDean]

@GithubProjects https://t.co/7gTQEv16rA

[Not Caleb Dean @FakeCalebDean]

Sheesh either @claudeai is slimming my tokens down or I've really just got that many agents doing work with loops they encourage

[Not Caleb Dean @FakeCalebDean]

My grok x Claude agent is impressive Claude and grok spin up agents swarm create build quickly and loop 🔁 until it gets done with no errors #agentic #claude #ai Claude spawns up to 16 agents to get work done

[Not Caleb Dean @FakeCalebDean]

https://t.co/k62M4MvIBI

[Not Caleb Dean @FakeCalebDean]

My grok x Claude agent is impressive Claude and grok spin up agents swarm create build quickly and loop 🔁 until it gets done with no errors #agentic #claude #ai Claude spawns up to 16 agents to get work done

[Not Caleb Dean @FakeCalebDean]

I'm tired of people pushing Bias narratives around ai and especially Fable 5. wait...something happened you aren't aware of. Yes 95% of people that used it had good intentions and the correct morals behind it. the 5% that got ahold of it that didn't

[Not Caleb Dean @FakeCalebDean]

🐋 WHALE ALERT ⭐⭐⭐⭐⭐

🏷️ A+ | COPY INSIDER BET
⏰ 16:53:32 | 06/19
👛 0x0346...52a5
🏟️ Will Morocco win on 2026-06-19?
📊 Yes (ML) | BUY | 58%
💰 $190,207

#fiba #morocco

https://t.co/Q2Xlc6mtBI

[Not Caleb Dean @FakeCalebDean]

@Nira_O351

[Not Caleb Dean @FakeCalebDean]

Big paper bet on polywave bets telegram channel #Morocco join now for free no strings attached!

[Not Caleb Dean @FakeCalebDean]

Big winner #usa with the big 2-0 win

[Not Caleb Dean @FakeCalebDean]

I’m gaining significant interest in what XAI/grok is doing with Codex… are the anthropic days over or just getting started? Customer loyalty is up in question. I have a hard time dealing with that thought process. It’s even hard to put into words. I want Fabel 5 back but I don’t want to show my ID when I have nothing to hide.

[Not Caleb Dean @FakeCalebDean]

Slow boring day in the market to be honest...same old news being repeated

[Not Caleb Dean @FakeCalebDean]

@ns123abc Count me in - there was a timeframe where one message I sent used 35% their customer service is a highly sophisticated automated email system and told me to get lost! Voice was never heard…no refund nor formal response but I still continue to use them 🤷🏽‍♂️

[Not Caleb Dean @FakeCalebDean]

One AI lab said its models are now building better models. Days later the government pulled its most powerful one within 72 hours of launch. The acceleration and the brakes landed in the same week. Nobody's ready for either. #AI #AGI