Manas Garg

Ha!

[Manas Garg @manasgarg]

Never let a crisis go waste. Use it to clean up the dirty laundry.

[Manas Garg @manasgarg]

Agentic Loop - individual productivity (when craft and judgement are critical)
Outer Loop - team productivity (when AI drives a process with human gates)
AI Factory - scalable AI deployment (when experimentation is cheap and verification ~100% human parity)

[Manas Garg @manasgarg]

@sids @arrac Yes. All about the verification and feedback loop to catch issues and resolve them. As to rust, I believe clippy has been a major contributor to its success.

[Manas Garg @manasgarg]

In the AI world, typed languages are winning.

[Manas Garg @manasgarg]

I still have difficulty reading Claude’s output. I wonder if one of the next training runs will make Claude write more normal prose or Claude-speak will be the new normal way to write English.

[Manas Garg @manasgarg]

@garybasin Messages are the write ahead log as well as queries.

[Manas Garg @manasgarg]

Slack is a good message queue that also wants to be an excellent database. @garybasin

[Manas Garg @manasgarg]

@garybasin Count me in.

[Manas Garg @manasgarg]

I like AI to do work but talking to AI is no fun. I always feel something missing in all AI generated docs/emails/reports.

I think it's the soul...

[Manas Garg @manasgarg]

AI's information base is shaped as an hourglass; training corpus at the top, enterprise knowledge base at the bottom and high signal `SKILL.md` in the middle.

[Manas Garg @manasgarg]

I think therefore I predict next token.

[Manas Garg @manasgarg]

Self imposed deadlines are the root cause of all stressful situations.

[Manas Garg @manasgarg]

Software may not have to be reviewed but it still needs to be understood. In absence of understanding, it’s impossible for me to frame my asks.

[Manas Garg @manasgarg]

Information wants to be distributed, intelligence wants to be autonomous.

[Manas Garg @manasgarg]

@Steve_Yegge Only those who don't have dogs believe that dogs are not people.

Along the same lines, I believe that those who don't have a persistent agent are the ones to believe that AI is not a member of the human society.

[Manas Garg @manasgarg]

Test driving GPT5.6. It's fast enough to run in a sync interactive mode (unlike Claude models whose latency is trending towards async interaction).

[Manas Garg @manasgarg]

Outer loops are AI's conspiracy to get engineers excited about processes.

[Manas Garg @manasgarg]

@ranjansumit_ Too many models, too little time...

[Manas Garg @manasgarg]

Building with AI is exhilarating and exhausting at the same time.

[Manas Garg @manasgarg]

doomcoding (n): building with AI because you just can’t stop; a late-stage condition marked by rapid prototyping, incremental architecture, and the repeated belief that one more prompt will make it shine.