Max Manuylov

Tech entrepreneur. AI builder

[Max Manuylov @MaxManuylov]

effort is a parameter to tune, not a dial to max out.

Start at medium. Move to high only where you’ve verified it helps.

[Max Manuylov @MaxManuylov]

Most common mistake I see with Claude: cranking effort to max “just to be safe.”

You’re not buying quality. You’re buying tokens.

Here’s the math on why max is almost never the right call 🧵

[Max Manuylov @MaxManuylov]

What I actually run:

• Notes, drafts, docs, routine → medium
• Real code, multi-file refactors → high
• Genuinely hard architecture → xhigh, rarely
• max → never once

[Max Manuylov @MaxManuylov]

no jailbreak — works because iOS trusts user-installed CAs and authy doesn't pin. best part: Claude debugged its own dead ends from the logs, like a stale expired mitmproxy CA throwing "SSL error" on the phone.

[Max Manuylov @MaxManuylov]

authy quietly killed the desktop app, the export, and the Raycast extension. mobile-only now. the only "supported" way off it: re-enroll 2FA on every account by hand — a pain that trips a security review on half of them.

[Max Manuylov @MaxManuylov]

it stood up a mitmproxy, walked me through the proxy + cert on the iPhone, and intercepted the encrypted token DB authy pulls on login. decrypted it with my backup password → every account as standard otpauth:// URIs. moved to Ente Auth.

[Max Manuylov @MaxManuylov]

the agentic coding discourse, 2026

[Thariq is on vacation @trq212]

https://t.co/R6exTuF7P8

[Max Manuylov @MaxManuylov]

If your Claude Code limits started draining faster since yesterday, the official security-guidance plugin is a likely culprit.
It used to be a simple deterministic hook. Now it spins up background review agents on every turn + commit — running Opus 4.7 by default.

[Max Manuylov @MaxManuylov]

@perplexity_ai Advanced? Without custom MCP support? Really?

[@levelsio @levelsio]

I made a trailer for the future of humanity (e/acc)

[Max Manuylov @MaxManuylov]

@OpenAI Has GPT appointed its nominal CEO?

[Max Manuylov @MaxManuylov]

@raycastapp Why is there no GPT-3.5 Turbo Instruct?

[Max Manuylov @MaxManuylov]

@cobular_ @hellokillian @raycastapp How about using the Raycast AI API instead of the OpenAI API?

[Max Manuylov @MaxManuylov]

Looking for ways to optimize instruction execution in gpt-3.5-turbo?
Pro tip: Custom instructions, like specifying a JSON format response, perform significantly better when incorporated into a user message rather than a system message. A subtle yet powerful tweak! #OpenAIapi

[Anil Chandra Naidu Matcha @matchaman11]

Stop writing shell commands manually 🔥🔥

Use Shell AI to generate queries in terminal 🦾

Just run pip install shell-ai to start usage

Github link below ↓