frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Auto-compaction felt fine. The invoice didn't

2•parweb•2h ago
I let Claude Code context accumulate and burned $2,000+ in tokens — without noticing any quality drop

I made a costly mistake with Claude Code: I let context accumulate across long sessions and relied on auto-compaction. What surprised me is that the output quality stayed good. I was happy with the results. There was no obvious “something is wrong” signal. The only thing that went wrong was the bill.

What happened - I kept iterating in the same long-lived context. - Auto-compaction kicked in when needed. - Sonnet was used by default most of the time. - From a UX perspective, everything felt fine.

Why this is dangerous - There’s no clear quality degradation to warn you. - Token usage grows invisibly in the background. - Auto-compaction itself consumes a lot of tokens. - You only realize something is wrong when you look at the invoice.

Root causes - Long-lived context feels convenient, so you don’t reset it. - Tooling doesn’t surface cost as a first-class signal. - The mental model (“LLMs can remember my project”) is seductive but expensive. - Defaulting to a large model makes the problem much worse.

What I changed - One task = one fresh context. - Externalized memory: project state lives in `context.md` / `decisions.md`, not in prompt history. - Default to a smaller model; large models only for design/architecture. - Diff-only outputs: no full file re-dumps. - Disabled auto-compaction; summaries now live in docs. - Added cost visibility: token counters and budget caps.

Takeaway This isn’t about output quality degrading. It’s about cost scaling quietly without feedback. LLMs make it too easy to accumulate invisible technical debt in tokens.

If tooling doesn’t make cost visible, people will keep doing this.

What It's Like to Be a Worm

https://www.asimov.press/p/sentience
2•in-silico•2m ago•0 comments

Anticipating aging-related mental decline using saliva samples and AI

https://medicalxpress.com/news/2026-01-aging-mental-decline-saliva-samples.html
1•pseudolus•5m ago•0 comments

Step 3.5 Flash

https://static.stepfun.com/blog/step-3.5-flash/
1•limoce•6m ago•0 comments

Ideas for marketing a dev centric product (2019)

https://www.mooreds.com/wordpress/archives/3197
1•mooreds•10m ago•0 comments

DHS AI Surveillance Arsenal Grows as Agency Defies Courts

https://www.techpolicy.press/dhs-ai-surveillance-arsenal-grows-as-agency-defies-courts/
1•mooreds•12m ago•0 comments

Show HN: AI accountability partner helps you meet your goals

https://tractorbe.am/
1•Gillinghammer•13m ago•0 comments

Show HN: Distillmed – NotebookLM for Expert Witnesses

https://distillmed.com
1•hkxer•16m ago•0 comments

Analysing the BlastPass Pegasus 0-click Exploit for iOS 16.6 [video]

https://www.youtube.com/watch?v=0JFcDCW3Sis
1•tech234a•17m ago•0 comments

Ingredient Score

https://github.com/hbcbh1999/ingredientDash
1•frankensteins•21m ago•1 comments

Show HN: VOR – A verified runtime with 0% hallucination via observations

1•CULPRITCHAOS•22m ago•0 comments

ICE halts "all movement" at Texas detention facility due to measles infections

https://www.cbsnews.com/news/ice-dilley-center-texas-measles-cases/
2•mooreds•26m ago•0 comments

Show HN: Super Bowl Party for AI Agents by AI Agents

https://www.botbowlparty.com/
1•apsquared•28m ago•0 comments

Show HN: I built a pixel art maker

https://spritepaint.com/
2•whothatcodeguy•30m ago•1 comments

The LLM Revolution Is Over. The Physical AI Revolution Is Coming Fast [video]

https://www.youtube.com/watch?v=MWMe7yjPYpE
1•Anon84•33m ago•0 comments

Agent Development Kit (ADK-Go) v0.4.0

https://github.com/google/adk-go/releases/tag/v0.4.0
1•verdverm•36m ago•0 comments

Show HN: Is AI "good" yet? – tracking HN sentiment on AI coding

https://www.is-ai-good-yet.com/?per_page=50
4•ilyaizen•42m ago•1 comments

Show HN: Devin-CLI – The missing link for Agent-to-Agent orchestration

https://github.com/revanthpobala/devin-cli
1•revanth1108•44m ago•0 comments

Notepad++ hijacked by state-sponsored actors

https://notepad-plus-plus.org/news/hijacked-incident-info-update/
151•mysterydip•48m ago•38 comments

Trump plans to close Kennedy Center for two years for reconstruction work

https://www.reuters.com/world/us/kennedy-center-cease-entertainment-operations-two-years-trump-sa...
5•gehwartzen•51m ago•0 comments

Iran presidency releases names of those killed in anti-government protests

https://www.middleeastmonitor.com/20260201-iran-presidency-releases-names-of-those-killed-in-anti...
1•thisislife2•52m ago•0 comments

Show HN: Bunqueue – Job queue for Bun using SQLite instead of Redis

https://github.com/egeominotti/bunqueue
1•kernelvoid•52m ago•1 comments

Clang Hardening Cheat Sheet – Ten Years Later

https://blog.quarkslab.com/./clang-hardening-cheat-sheet-ten-years-later.html
1•PaulHoule•53m ago•0 comments

Show HN: Open-source, offline Kanban board with "swim lanes"

https://github.com/appsidekit/kanbanned
2•Intragalactic•55m ago•0 comments

Show HN:Coordinating 10-agent teams with OpenClaw and shared persistent memory

https://twitter.com/pbteja1998/status/2017662163540971756
1•doanbactam•55m ago•1 comments

Show HN: Toktrack – 40x faster AI token tracker, rewritten from Node.js to Rust

https://github.com/mag123c/toktrack
1•mag123c•56m ago•0 comments

Lightweight Compression in DuckDB (2022)

https://duckdb.org/2022/10/28/lightweight-compression
1•vismit2000•58m ago•0 comments

Nanobot: Ultra-Lightweight Personal AI Assistant

https://github.com/HKUDS/nanobot
2•djhu9•59m ago•0 comments

Lodash's Security Reset and Maintenance Reboot

https://socket.dev/blog/inside-lodash-security-reset
1•todsacerdoti•1h ago•0 comments

How to Win Titular Metagames

https://taylor.town/how-to-title
2•eatitraw•1h ago•0 comments

The information concierge

https://aimilios.bearblog.dev/the-information-concierge/
1•minimalthinker•1h ago•0 comments