frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: An "Evidence Loop" for steering the agents what to build

https://www.dipio.ai
1•jkalichman•1h ago
Fellow builders, my name is Julian - I'm a PhD candidate in Behavioural Science at LSE, where I focus on Human-AI Alignment (specifically the field of Machine Psychology), and we built Dipio AI around a problem I think is becoming more important as coding agents get better: implementation is getting dramatically cheaper, but knowing what is worth building isn't.

Dipio applies behavioral science methodologies to AI-led user research, turning evidence from user conversations and behavior into agent-native specs (e.g., spec-kit, etc.). Your coding harness can access them directly through Dipio MCP, or through the platform as an evidence-backed, ready-to-implement feature specs.

I wrote more about the idea we call the “Evidence Loop” here: https://www.dipio.ai/blog/evidence-loop

I’d particularly appreciate criticism of the underlying thesis: are we solving the "how to build" problem much faster than the "what to build" problem? Also, any criticism in general is welcomed, don't hold yourself - I have a hard skin, and most importantly I value a lot the feedback from such a great community like this one!

Comments

jkalichman•53m ago
A bit more context on why we built this:

A lot of the recent signals point in the same direction: back in Dec 2025, Boris Cherny said 100% of his Claude Code contributions "over the previous 30 days" were written by Claude, and as Garry Tan recently mentioned - YC has already seen companies with 95% AI-generated codebases.

We’re getting very good at the "how".

But most of the systems around software development still optimize for shipping faster and more reliably. They don’t really answer the more basic question: was this worth building?

That’s the idea behind what we call the Evidence Loop:

real users (or synthetic copy of it) → evidence → spec → agents build → observe what happens → start the loop again.

We’re also researching Synthetic Twins. We’ve published peer-reviewed work (currently in press). I don’t think synthetics should replace real users; the interesting part is using them to explore hypotheses cheaply, then bringing those hypotheses back to real users and real behaviour.

Longer term, the bet is simple: agents shouldn’t just get a ticket saying “build X.” They should have access to the evidence explaining why X should exist at all.

Thank you!

Julian

DM's are welcomed: https://www.linkedin.com/in/j16h/

Terence Tao creates his first meme

https://mathstodon.xyz/@tao/117068266071803252
1•u1hcw9nx•2m ago•0 comments

London still dominates Britain's datacenter map, but the regions are powering up

https://www.theregister.com/systems/2026/08/10/london-still-dominates-britains-datacenter-map-but...
1•01-_-•3m ago•0 comments

Anthropic's New Tool Could Expose AI-Written Novels

https://www.businessinsider.com/anthropic-watermarking-feature-stops-undetected-ai-generated-writ...
1•01-_-•4m ago•0 comments

CortexBrain: A monitoring pipeline that integrates with your AI agents

https://github.com/CortexFlow/CortexBrain
1•LorenzoTettaman•5m ago•1 comments

The US hired 2k gamers to fix the air traffic control crisis

https://www.polygon.com/fortnite-united-states-air-traffic-control/
1•croes•5m ago•0 comments

First measurement of antineutrinos from spent nuclear fuel

https://www.mpi-hd.mpg.de/mpi/en/public-relations/news/news-item/first-measurement-of-antineutrin...
1•croes•12m ago•0 comments

A Simple TODO Application

https://codemadness.org/todo-application.html
2•hskimse•13m ago•0 comments

Scope and Lifetime Restrictions in Swift

https://github.com/rjmccall/swift-evolution/blob/scope-restrictions/scope-restriction-models.md
1•matt_d•20m ago•1 comments

Unikernel

https://en.wikipedia.org/wiki/Unikernel
1•tosh•21m ago•0 comments

Australian Wealth Tracker

https://auravest.ai
1•mozzinator•22m ago•1 comments

Excited to see the results of the agent memory leaderboard

https://x.com/AgentMemoryL
1•IreneAI•23m ago•0 comments

An oil field became California's first carbon vault

https://calmatters.org/environment/2026/08/california-resources-corporation-carbon-terra-vault/
1•geox•24m ago•0 comments

Instant Sudoku

https://sudokuaz.com/
2•Boryto•26m ago•0 comments

The pain of integration with the Reddit API

https://growomat.com/blog/reddit-ads-api-integration-lessons
1•nickdefo•26m ago•0 comments

Show HN: XIRP Agentic Harness by Spotify

https://xirp.spotify.com/
1•slowey•28m ago•0 comments

Scientists find 200k barrels of radioactive waste in the ocean

https://geekspin.co/scientists-find-200000-barrels-of-radioactive-waste/
2•_tk_•29m ago•1 comments

helppeer.ai

https://helppeer.ai/llms.txt
1•tosh•31m ago•0 comments

The mathematical beauty of hyperbezier curves

https://linebender.org/blog/hyperbezier/
3•coffeeaddict1•33m ago•0 comments

Stealing Reasoning Traces from Proprietary LLM APIs

https://arxiv.org/abs/2608.09867
1•sbulaev•35m ago•0 comments

Notes from China, Q3 2026

https://milan.cvitkovic.net/china_notes_q3_2026/
1•ridiculous_leke•36m ago•0 comments

AI Slop Backlash Is Starting to Change Platform Policy

https://ai-updates.net/ai-slop-backlash-changing-platform-policy/
2•ashurandi•36m ago•0 comments

Ask HN: Anyone have solution to Opus verbosity in Claude Code?

2•firemelt•37m ago•0 comments

The mathematical physics of rainbows and glories(2001) [pdf]

https://tlakoba.w3.uvm.edu/AppliedUGMath/auxpaper_rainbow_glory_review.pdf
1•num42•38m ago•0 comments

Local models will not win

https://www.seangoedecke.com/local-models-will-not-win/?trk=feed_main-feed-card_feed-article-content
3•sylvainkalache•39m ago•1 comments

AI professors are negotiating the new realities of academic research

https://www.technologyreview.com/2026/08/10/1141597/ai-professors-are-negotiating-the-new-realiti...
4•joozio•40m ago•0 comments

Invoice Software for Freelancers – InvoiceAce v2 Released

https://invoiceace.app
1•nimsarajay•41m ago•0 comments

The Water Footprint of AI

https://doi.org/10.1016/j.watres.2026.125866
3•MASNeo•41m ago•0 comments

Looking for free TTS? Try this one

https://neuronai.uz/en/tts/
1•yutabek•42m ago•0 comments

Sign of the Times: An Account of the First (In the USA) Tech Workers' Strike

https://longhaulmag.com/first-tech-workers-strike/
3•chobeat•42m ago•0 comments

Show HN: Backpressure – a load simulator for system design and LLM serving

https://backpressure.systems
1•leocardz_•44m ago•0 comments