_private/today.md

star
Table of Contents

claim reset, needs to reset confirm state

https://candidate-staging.qwestly.co/api/admin/user-info

https://candidate-git-feat-qwestly-agent-qwestly.vercel.app/admin/qwestly-agent https://qwestly-agent.vercel.app/api/health

https://blog.kilo.ai/p/we-tested-deepseek-v4-pro-and-flash

Re-architecting Qwestly

  • lazy load previous messages. put a cap
  • knowledge base about what qwestly can do

we have a lot of prompts, service files, etc. adding a new one is cumbersome, need api endpoints, storage

  • need agentic system first, but how can we spin this up easier
  • everything is tool calls in the agent loop
  • tool figures out intent -> maps to prompts/skills
    • tools call these prompts (still in langsmith? or stored local)
    • for simplicity, we should have 2 general models - flash and pro (for chat and complex tasks)
    • special prompts can have its own override, but most should use the previous 2
    • lets us upgrade easier in batches
    • need some system to dictate where output goes / where it gets stored & retrieved with proper auth/permission
  • new agentic feature would just spin up a new tool

Question: how to maintain memory

  • chatbot like open-claw creates daily & session, short term + long term.
  • we need memory types rather short vs long: achievements

Data types to figure out:

  • structured fields
  • memory - unstructured but summarized / synthesized
  • RAG

Bite-sized

Section 5 (Data Modeling)** — the embedded subdocument map approach vs. separate collections is a design decision worth validating

  • Section 7 (Implementation Details) — the file plan has ~30 files; we could collapse or expand certain modules
  • Section 8 (Proportionality) — the "knowledge agent" concept in the pipeline might be overkill for v1 and could be simplified
  • Section 10 (Risks & Open Questions) — several open questions are tagged to Adam; we may want to separate pre-implementation blockers from nice-to-haves
  • Section 3 (Requirements) — acceptance criteria could be more testable/measurable

Today

Lao Sze Chuan https://maps.app.goo.gl/AqNpaqoqHnfCCEh4A https://candidate-git-chore-employment-stints-qwestly.vercel.app/admin/employment-stints

  • David needs health insurrance in July

https://github.com/EnotionZ/max-chrome-extension

  • no way to return to call
  • claim account

https://github.com/deepseek-ai/awesome-deepseek-agent

Sprint planning

  • Asana MCP

  • HM fast onboarding is on staging

  • Networking feture

    • finalizing copy
    • popular companies

  • Voice debug

    • client-side, made some fixes to audio selection and storing local stream
    • approved eval pipeline plan

  • Bite-sized interview

  • Recruiting efforts that need tooling

One person... noted that Mr. Trump had a habit of confusing tactical advice from General Caine with strategic counsel. In practice, that meant the general might warn in one breadth about the difficulties of one aspect of the operation, then in the next note that the US had an essentially unlimited supply of cheap, precision-guided bombs and could strike Iran for weeks once it achieved air superiority. To the chairman, these were separate observations, but Mr. Trump appeared to think that the scond most likely canceled out the first.

you’re trying to solve something that’s never been solved before, there’s tremendous technological or intellectual gaps in our knowledge that maybe having a less formal culture and allowing people to go off on tangents ... [is] exactly the right thing to do

In prolonged conflicts, maintaining interceptor stocks and coordinating defense across multiple sites becomes a major strategic challenge. “Once you get into repeated raids, mixed salvos, and long-duration drone pressure, the limiting factor becomes magazine depth, resupply speed. and the economics of using very expensive interceptors against cheap, persistent threats,”

“It’s a bit of a dilemma for incumbents,” he told me. “Can you express empathy for people’s economic anxieties without seeming to concede that all is not rosy on your watch?”

Launch announcements

  1. what did we launch?
  2. why is it important?
  3. how does it work?
  4. how will this be reported?