Mr. Buzzoni

AI dev • co-founder @polynternet
• Building in progress

[Mr. Buzzoni @Mr_Buzzoni]

Jev + Claude Opus 5.5 is certified f*cking gold... i can't understand why everyone isn't wiring this up yet

the forks alone drop from about $110 a day to $4 once Jev takes them over. right now it's the best way i know to run Opus

the jobs Jev takes off Opus:
-> deciding which project notes are worth loading before the context fills up
-> routing the easy steps to a faster worker
-> choosing how to recover when a tool breaks, so nothing retries on autopilot
-> running the narrow checks first and saving the full test suite for last

Opus keeps the hard reasoning. Jev only picks between options the harness already laid out and tested

each pick lands in about 100 ms, and Jev's output tokens cost nothing

Jev engineering in a sentence: let the expensive model think and the cheap one choose

$106 a day saved on identical work is the kind of win that comes up at your next salary review

read the paper first, then the full Jev guide below

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/SxHnsCMK6g

[Mr. Buzzoni @Mr_Buzzoni]

SpaceXAI put out a 12-page PDF on the best GrokBot use cases - read it and you'll use GrokBot better than 95% of people

what the bot is, which use cases work today, and how to deploy it without burning trust:

step 1 -> every working use case has one shape: read, check, apply a rule, write a draft. if your task fits, the bot can take it

step 2 -> five checks before handoff: repeatable, cross-system, rule-based, draft-first, cheap to get wrong. fails one? add a gate or wait

step 3 -> start with sales. 36 LinkedIn drafts queued overnight, none sent without review. no team has more agent-ready work

step 4 -> engineering gets bug repro, CI/CD watch and dependency audits. the bot picks up the ticket, hits staging and hands back a repro pack

step 5 -> operations hands over inbox, expenses and travel. the bot logs into portals that have no API, and finance gets 8-12 hours back at month-end

step 6 -> marketing owns 13 of the 56 roles: paid media, content calendars, competitor watch. all overnight, all draft-first

step 7 -> the approval line: every documented use case stops before external action. nothing gets sent, booked or paid without you

step 8 -> roll out over four weeks: watch, review everything, review flags, then a scheduled routine. skip a week and the risk stacks up

step 9 -> the fleet: a Chief of Staff bot on top, specialists below, bots messaging each other. you stop being the middleman

step 10 -> rank tasks by the cost of a mistake. the bot nails 49 in a row, then does something strange on the 50th with full confidence. keep a human at every exit

the result: 56 roles, one playbook. the bot takes the 80% that repeats, you keep the 20% that matters

then read how to build a 24/7 GrokBot agent team in the article below

[Mr. Buzzoni @Mr_Buzzoni]

this is absolute f*cking gold

a SpaceXAI engineer shared his prompt for a 30+ agent GrokBot team, built on Lauren Tan's stack

one paste and GrokBot builds the whole team for you

the full roadmap for running it is in the article below

[Mr. Buzzoni @Mr_Buzzoni]

SpaceXAI put out a 12-page PDF on the best GrokBot use cases - read it and you'll use GrokBot better than 95% of people

what the bot is, which use cases work today, and how to deploy it without burning trust:

step 1 -> every working use case has one shape: read, check, apply a rule, write a draft. if your task fits, the bot can take it

step 2 -> five checks before handoff: repeatable, cross-system, rule-based, draft-first, cheap to get wrong. fails one? add a gate or wait

step 3 -> start with sales. 36 LinkedIn drafts queued overnight, none sent without review. no team has more agent-ready work

step 4 -> engineering gets bug repro, CI/CD watch and dependency audits. the bot picks up the ticket, hits staging and hands back a repro pack

step 5 -> operations hands over inbox, expenses and travel. the bot logs into portals that have no API, and finance gets 8-12 hours back at month-end

step 6 -> marketing owns 13 of the 56 roles: paid media, content calendars, competitor watch. all overnight, all draft-first

step 7 -> the approval line: every documented use case stops before external action. nothing gets sent, booked or paid without you

step 8 -> roll out over four weeks: watch, review everything, review flags, then a scheduled routine. skip a week and the risk stacks up

step 9 -> the fleet: a Chief of Staff bot on top, specialists below, bots messaging each other. you stop being the middleman

step 10 -> rank tasks by the cost of a mistake. the bot nails 49 in a row, then does something strange on the 50th with full confidence. keep a human at every exit

the result: 56 roles, one playbook. the bot takes the 80% that repeats, you keep the 20% that matters

then read how to build a 24/7 GrokBot agent team in the article below

[Gipp 🦅 @gippp69]

https://t.co/gTEKJ3i4XU

[Mr. Buzzoni @Mr_Buzzoni]

this is absolute f*cking gold

a SpaceXAI engineer shared his prompt for a 30+ agent GrokBot team, built on Lauren Tan's stack

one paste and GrokBot builds the whole team for you

the full roadmap for running it is in the article below

[Iron Giant @irongiantXBT]

https://t.co/PCyBGgl3RS

[Mr. Buzzoni @Mr_Buzzoni]

I can't believe this is f*cking free

how to build your first AI agent (full guide, with a Jev decision layer)

my first agent ate two weeks of my life. this guide gets you to a working one in a single afternoon

in the right hands it lets one builder ship what used to take a whole team:

[Mr. Buzzoni @Mr_Buzzoni]

Claude Code tip: once Opus 5.5 is your main model, stop letting your Fable 5.1 quota go to waste

put it on call with /advisor
↳ https://t.co/JQWFFf8SXx

run /advisor fable
(on some plans, run /model fable once first to accept usage-credit billing)

Opus 5.5 keeps working
Fable 5.1 reads the whole session, every tool call included, and steps in at three moments:

-> before a plan: is this the right approach?
-> when the same error keeps coming back: am I digging in the wrong place?
-> before "done": what did I miss?

press Ctrl+O to read what it told Opus

the clever part: toggling /advisor leaves the Opus prompt cache intact, so you only pay for the few moments Fable actually reads the session

and when the advice clashes with what's in the files, Opus flags the conflict and keeps going on the evidence

it's the core idea of Jev engineering: the expensive model stays out of routine steps and only shows up where the outcome can change

• the full setup
> Opus 5.5 on high runs the main session
> subagent one reads code
> subagent two edits and runs tests
> subagent three looks up docs
> all three on medium, all three inherit the advisor
> Fable 5.1 on call

hand the tree and this prompt to Claude Code 👇

"Set up my Claude Code to match this tree:

1. Reuse fitting subagents from ~/.claude/agents and .claude/agents.

> Propose new ones only for missing roles
> Set each to model: opus, effort: medium
> Leave any that set a different model alone and list them

2. Set main session effort to high via effortLevel in ~/.claude/settings.json, and confirm advisorModel is set to fable there

3. Check for env vars that disable the advisor (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, anything that stops flag fetching) and CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort. Report them and leave them as they are

4. Add a rule to ~/.claude/CLAUDE.md: consult the advisor before a big plan, when an error repeats, and before calling a long task done

Show me the changes first. Don't edit files yet."

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/hIl5jyrAQz

[Mr. Buzzoni @Mr_Buzzoni]

I can't believe this is f*cking free

how to build your first AI agent (full guide, with a Jev decision layer)

my first agent ate two weeks of my life. this guide gets you to a working one in a single afternoon

in the right hands it lets one builder ship what used to take a whole team:

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/SxHnsCMK6g

[Annatar.md @AnnatarXBT]

this is pure f*cking gold for anyone running coding agents

Jev founder Diogo Amogo wrote a PDF on building a Jev harness

what their tests show:
> up to 200x faster
> up to 400x cheaper

the model hasn't been the bottleneck for a while
speed and cost both come from the harness you wrap around it

• how to use it
> give this PDF and the article below to Claude Code or Codex
> ask it to rebuild its own setup around Jev

• how it turns into a promotion
> pick the agent workflow your team pays the most for
> rebuild it in one evening
> bring the before and after bill to your manager

one evening of Jev engineering, and next week you're the one running a harness most teams haven't built yet

[Mr. Buzzoni @Mr_Buzzoni]

Claude tip: one prompt from an Anthropic engineer turns Opus 5.5 and Fable 5.1 into a full agent team

the output is f*cking unreal

paste it into either model and thank me later

the full roadmap with examples is below

[Mr. Buzzoni @Mr_Buzzoni]

Claude tip: one prompt from an Anthropic engineer turns Opus 5.5 and Fable 5.1 into a full agent team

the output is f*cking unreal

paste it into either model and thank me later

the full roadmap with examples is below

[Kirill @kirillk_web3]

https://t.co/BSAEmC5oKL

[Mr. Buzzoni @Mr_Buzzoni]

@kirillk_web3 glad it was useful enough to share

[Mr. Buzzoni @Mr_Buzzoni]

ok this is f*cking gold for students

how to get Wispr Flow Pro free for a whole year with a student email

everything in Pro for $0 instead of $144 a year, no credit card

how to claim it:

1. use your student email, or google a free .edu email if you don't have one

2. sign up for Wispr Flow with it https://t.co/U67fxomd3j

3. download the desktop app

4. open Settings -> Plans and Billing and claim the Student plan

what you get:
> unlimited words every week
> command mode to edit text by voice
> learns your writing style
> 100+ languages

pair it with Claude Code and you start speaking your prompts instead of typing them

no card on file means there's nothing to cancel when the year runs out

[Mr. Buzzoni @Mr_Buzzoni]

this article will cut your AI bill by 90% with Jev engineering

same 100,000 messages a month
$7,950 -> $773

Jev sorts the inbox in one call, and Kimi K3 only writes the replies that need real thinking

it's the layer after harness and loop engineering, and it works on anything your AI agents touch a hundred times a day:
> support tickets
> inbound leads
> Claude Code tool calls

bring that number to your next budget meeting and the raise conversation starts itself

the full breakdown is below:

[leopardracer @leopardracer]

one tiny sorter model just made your AI support budget look like a f*cking scam

$7,950 a month down to $773, same 100,000 messages, same quality

the big model only wakes up when it actually matters

set it up once and stop paying premium prices ↓ https://t.co/YOL5ST3iW2

[Mr. Buzzoni @Mr_Buzzoni]

this is the f*cking cheat code for anyone still stuck on their first AI agent

how to build your first AI agent (full walkthrough)

a year ago this would have saved me two weeks of trial and error. my first agent would have shipped before dinner

in the right hands it turns one person into a small dev team:

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/mGAa1zICki

[Mr. Buzzoni @Mr_Buzzoni]

this is pure f*cking gold

Andrej Karpathy landed at Anthropic 20 weeks ago

two things are already public, and together they're the whole playbook:

> Anthropic's own knowledge graph cookbook
> Karpathy's autoresearch loop: 700 experiments in 2 days, 20 optimizations found with nobody steering

stack them and you get graph engineering on top of a loop

i wired the graph approach into my own setup. the first reply already felt different

Claude dropped the canned answer and reasoned through the problem step by step

the loop runs the experiments. the graph decides what's worth remembering

read the cookbook first, then the article below

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/buH2nYDL29

[Mr. Buzzoni @Mr_Buzzoni]

Jev + Claude Opus 5.5 is a f*cking goldmine... i can't understand why everyone isn't wiring this up yet

the forks alone drop from about $110 a day to $4 once Jev takes them over. right now it's the best way i know to run Opus

where Jev steps in:
-> ranks which project notes matter before the context loads
-> passes the easy steps to a faster worker
-> chooses a recovery path when a tool fails, so nothing retries blind
-> runs the targeted checks before the full test suite

Opus does the hard reasoning. Jev chooses between options the harness already built and checked

every choice comes back with a probability, so the harness can set a confidence floor and push anything below it back up to Opus

Jev engineering fits in one line: the expensive model reasons, the cheap one decides

$106 a day saved on the same work ends up in your performance review

start with the paper, then the full Jev guide below

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/hIl5jyrAQz

[Mr. Buzzoni @Mr_Buzzoni]

Claude Haiku 5.5 reads a million lines of code for about a dollar

and it's smart enough to put them to work:

> scan your whole repo and hand Opus 5.5 only the files that matter
> summarize every PR, diff and failing CI log
> write and run your database queries
> compact long coding sessions so the big model stays cheap
> click through a browser and finish multi-step tasks on its own

the setup worth stealing: Opus 5.5 plans, Sonnet 5.5 builds, Haiku 5.5 runs the hundreds of small calls in between

one dial sets how hard it thinks on each task, so cheap jobs stay cheap

Max and Team plans now come with monthly API credits to build all of this: $100, $200 or up to $500

the small model is now the one you can leave running all day

[Claude @claudeai]

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.

On average, it costs around 75% less to run than Claude Haiku 4.5.

[Mr. Buzzoni @Mr_Buzzoni]

OPUS 5.5 + ZETTELKASTEN FOR CODE = AN OBSIDIAN-STYLE SECOND BRAIN FOR YOUR REPO

your coding agent burns half its session on "read this file", "now read that one", "grep the whole repo"

codebase-memory fixes that

it indexes the Linux kernel, 28M lines, in 3 minutes
your repo takes seconds

one index turns the whole repo into a single graph
every function is a note, every call and dependency is a link

one query to that graph replaces dozens of grep and read cycles, so Opus 5.5 spends its tokens on thinking

benchmarked on 31 real repos:
-> 10x fewer tokens on structural queries
-> 83% answer quality on complex tasks
-> 2.1x fewer tool calls

paste this into Claude Code or Codex 👇

"Set up codebase-memory on this repo for me and show me the map.

Install it from https://t.co/z2gu1GGMnh:
curl -fsSL https://t.co/M76lL2dQ9H | bash -s -- --ui

Then restart, index this project, and open localhost:9749 so I can see the graph."

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/zujwYXVfAU

[Mr. Buzzoni @Mr_Buzzoni]

Claude Opus 5.5 with this prompt is a f*cking cheat code

paste it in and Opus 5.5 designs the whole system for you:
> the agents and what each one owns
> the prompt each agent runs on
> the workflow, handoffs and stop conditions between them

Claude Code tip: run it in plan mode first, cut every agent without a clear output, then let it build

grab it before it becomes everyone's default playbook:

[Mr. Buzzoni @Mr_Buzzoni]

https://t.co/ge5Zmq3pPN