Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Show HN: SQLBraid – Write SQL directly in TypeScript without a query-builder DSL(github.com/clickin ↗)
    discuss
  2. The Roadmap to Mastering LLM Inference Optimization(machinelearningmastery.com ↗)
    discuss
  3. Will new grid rules put the future of UK datacentres at risk?(computerweekly.com ↗)
    discuss
  4. He was probably right: we should be confident in our beliefs, but never certain(aeon.co ↗)
    1comments
  5. GeoJev, ask questions about the world and get its heatmaps(geojev.vercel.app ↗)
    discuss
  6. Do people prefer traditional architecture?(worksinprogress.co ↗)
    discuss
  7. Want to have a 24/7 colleague for your recruitment?(sourcegeek.com ↗)
    discuss
  8. Anthropic Principle(wikipedia.org ↗)
    1comments
  9. Big Bang(wikipedia.org ↗)
    discuss
  10. Deterministic Kittens: Fun with Qwen-Image-2.1 on an M2 MacBook Pro(reddit.com ↗)
    discuss
  11. Rolling with the Punches(debarshibasak.github.io ↗)
    discuss
  12. Day One Support for the Duo(timleland.com ↗)
    discuss
  13. The End Of Upward Mobility – AI is coming for the meritocracy(noemamag.com ↗)
    discuss
  14. BugBash'26 Keynote: We won, what now?(muratbuffalo.blogspot.com ↗)
    discuss
  15. Virtual Worlds(worksinprogress.co ↗)
    discuss
  16. The toilet took five minutes. My phone took the rest(salari-dev.medium.com ↗)
    discuss
  17. The Conveniences I Used to Write a Book About Inconvenience(lithub.com ↗)
    discuss
  18. ProviderCallbackHost is an allow-list, not a destination(nkap.dev ↗)
    discuss
  19. CSE 452: Google's Introduction to Distributed System Design(washington.edu ↗)
    discuss
  20. Mvndy (Comic)(mvndy.cl ↗)
    discuss
  21. Archive.yesterday: an empirical study of the Streisand effect(gyrovague.com ↗)
    1comments
  22. Tired of Astra eating all my credits. DotSwarm to the Rescue(github.com/druidia-bot ↗)
    discuss
  23. Probes into Toronto's shooter-for-hire network thwarted by impenetrable phone(theglobeandmail.com ↗)
    1comments
  24. Sifty, free text classification with one URL and no API key(sifty.dev ↗)
    1comments
  25. Ask HN: How do you guys stay on top of the code that is generated?
    3comments
  26. Jev Use Cases Tested: Where This Decision-Only AI Fits(mindstudio.ai ↗)
    discuss
  27. PDF Artifacts: What they are and why they matter for Accessibility(pdf4wcag.com ↗)
    4comments
  28. SoftBank Seeks Over $11 Billion in Junk Bonds for OpenAI Bet(bloomberg.com ↗)
    1comments
  29. Show HN: See what ChatGPT, Claude and Gemini say about your business(kelriva.ai ↗)
    discuss
  30. The Coal Conquest – There was no shortage of trees(ageofinvention.xyz ↗)
    1comments

Sleeper Service: Agents as a Service. One agent. One task. A thousand of them

1 pointsby 58m agogithub.com
1 comments
58m agoHN ↗

I have a few core observations about work that influence how I approach AI:

1. There is one-time work (development, personal assistant) where the input and output are net new. There is also back office work (billing, FP&A, questions about data) that is quite repetitive--working within existing datasets.

2. Back office work exists within SaaS products all of which are including AI within their platforms. Those SaaS products are often integrated with each other.

3. Business logic needs to be transferred between systems in order for each SaaS AI to be accurate. (Asking a Salesforce AI and a Xero AI "What is our revenue?" may get you different answers if either system has custom logic that defines any input to that question differently.)

Based on this, I believe that the optimal path forward for back office work is:

1. Centralize agents (business logic) across an organization

2. Decouple agents from orchestration

Agents that exist at the end of an API call--which can be from a SaaS, Airflow, vibe-coded app, or model harness--are a better answer than platform specific AI. They provide a centralized repository for business logic while reducing vendor reliance (and lock in). Additionally, you get better observability and auditabilty.

Sleeper Service is the result of this thinking--an enterprise AI framework.

Larger write up of my thought process behind Sleeper Service here: https://zero2data.substack.com/p/agents-as-a-service