19:[[47656335,{"by":"jer0me","descendants":61,"id":47656335,"kids":[47659598,47658579,47660276,47658764,47661498,47659081,47658381,47659787,47658385,47685279,47659148,47671475,47684685,47658318,47660819,47660556,47659267,47658803,47658208,47656517,47658906,47660026,47656475,47659908],"score":63,"time":1775442801,"title":"Does coding with LLMs mean more microservices?","type":"story","url":"https://ben.page/microservices"}],[47655408,{"by":"armanified","descendants":134,"id":47655408,"kids":[47658988,47660830,47658540,47656472,47657431,47659731,47660517,47656263,47657381,47657440,47656073,47658470,47656569,47661663,47661611,47661880,47664221,47657270,47657236,47656768,47661474,47659214,47666522,47656684,47657319,47664234,47659509,47659103,47657734,47659428,47663129,47659002,47658519,47661348,47662856,47656191,47657643,47658155,47662556,47658488,47660754,47661009,47676800,47658856,47657245,47660163,47656594,47671512,47658213,47793780,47741102,47656209,47662873,47662844,47660950,47661054,47656699,47662666,47656884,47661916,47657516,47660994,47657966,47657975,47667374,47664429,47659709,47659976,47656360,47657037,47658197,47658524,47658025,47657218,47661931,47659792,47657214,47656687,47656815,47659993,47657756,47657185,47658521],"score":915,"text":"Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food.

Fork it and swap the personality for your own character.","time":1775434812,"title":"Show HN: I built a tiny LLM to demystify how language models work","type":"story","url":"https://github.com/arman-bd/guppylm"}],[47598766,{"by":"novachen","descendants":1,"id":47598766,"kids":[47600130],"score":2,"text":"A CLI for fetching data from social platforms (X, Reddit, HN, GitHub, LinkedIn, Product Hunt, etc.) with compressed JSON output. Built because raw GraphQL responses from X are ~3,900 tokens for 10 tweets — this gets it down to ~666 tokens.

Features:\n- 15+ platforms supported\n- Three-tier auth (public API -> cookie -> OAuth)\n- Structured output designed for LLM context efficiency\n- Handles X's rotating query IDs and x-client-transaction-id

Built on top of jackwener's opencli/twitter-cli/rdt-cli ecosystem. Open source, PRs welcome.","time":1775036631,"title":"Show HN: Crossmind-CLI – CLI for X, Reddit, HN API with compact JSON output","type":"story","url":"https://github.com/cross-mind/crossmind-cli"}],[47371289,{"by":"xodn348","descendants":6,"id":47371289,"kids":[47384093,47374465,47377710,47418666,47380880,47371611,47392658],"score":4,"text":"$1a","time":1773443764,"title":"Ask HN: Has anyone built an AI agent that spends real money?","type":"story"}],[47380295,{"by":"jimsojim","descendants":114,"id":47380295,"kids":[47381773,47381371,47384439,47382088,47384637,47381724,47387375,47382331,47391000,47383725,47381462,47383838,47381605,47381636,47383180,47382189,47381858,47384992,47383791,47383953,47388610,47381303,47383863,47381086,47384994,47391657,47384792,47382555,47390267,47382453,47386563],"score":234,"time":1773516940,"title":"Marketing for Founders","type":"story","url":"https://github.com/EdoStra/Marketing-for-Founders"}],[47356614,{"by":"pauletienney","descendants":9,"id":47356614,"kids":[47357128,47382609,47362386,47374470,47359477,47356676,47366843,47357954,47359293],"score":15,"text":"I used to seek focus and concentration while coding. It was not always easy to reach this flow state but I knew it was possible.

I am now using agentic coding quite a lot. The honeymoon is finishing and I am starting to dislike some facets of it. I think the main setback is the rythm.

Writing some specs/prompts, launching the agent, confirming quite atomic actions and waiting 10 to 30 seconds until the next question/confirmation. Those very small wait times do not let me reach a concentration state.

I feel I am hovering the code. I am not deep into it as I used to be.

Do you feel the same? Did you find a way to change this?","time":1773347388,"title":"Ask HN: How do you cope with the broken rythm of agentic coding?","type":"story"}],[47332177,{"by":"Barathkanna","descendants":23,"id":47332177,"kids":[47362375,47418649,47395797,47346646,47395259,47343952,47333592,47335341,47332359,47332282,47353587,47336644,47332545],"score":5,"text":"I’ve been experimenting with agent-based features and one thing that surprised me is how hard it is to estimate API costs.

A single user action can trigger anywhere from a few to dozens of LLM calls (tool use, retries, reasoning steps), and with token-based pricing the cost can vary a lot.

How are builders here planning for this when pricing their SaaS?

Are you just padding margins, limiting usage, or building internal cost tracking?\nAlso curious, would a service that offers predictable pricing for AI APIs (like a fixed subscription cost) actually be useful for people building agentic workflows?","time":1773209195,"title":"Ask HN: How are people forecasting AI API costs for agent workflows?","type":"story"}],[47222844,{"by":"starfun","descendants":1,"id":47222844,"kids":[47361895,47397403],"score":2,"text":"$1b","time":1772479566,"title":"Ask HN: Using OpenClaw for marketing: worth it or overhyped?","type":"story"}]]

1.	Show HN: Crossmind-CLI – CLI for X, Reddit, HN API with compact JSON output(github.com) 2 points by novachen 47 days ago \| 1 comment
2.	1 points by novachen 47 days ago \| discuss
3.	1 points by novachen 141 days ago \| discuss

user:	novachen
created:	December 27, 2025
karma:	12