Digestavi.nyc

Avi's weekly digest, 2026

The week of Oct 10 to 17.

Oct 10 to Oct 17, 2026. 6 bookmarks saved this week, sorted into 3 topics.

01

AI agents and the software factory

4 bookmarks

  1. How employment is changing at Anthropic

    anthropic.comSaved Oct 17#1

    Anthropic's own research on adopting AI: agent compute hours now exceed human hours 6-to-1, junior technical hiring is declining as Claude-enabled coding spreads, and a new bottleneck in human attention is emerging as productivity per employee rises.

    Why it matters. you're running an OpenClaw-based software factory and deciding how much of the engineering org agents can actually carry; this is the sharpest real-world data point on agent-vs-human output and the "attention bottleneck" framing is directly relevant to your review-layer design.

  2. PostHog (@posthog) on X

    x.comSaved Oct 16#2

    Context engineering is the design of what an AI agent should read, know, and do — instructions, tools, data, memory — and when to use them. PostHog's Wizard is an autonomous onboarding assistant that installs PostHog into a codebase in minutes, built on a "context mill" pipeline of docs, example apps, and hand-written instructions.

    Why it matters. context engineering is exactly the discipline behind your Innovent Gateway routing and the software factory's agent prompts; the PostHog Wizard is a concrete case study of packaging instructions, tools, and memory into a working installer agent.

  3. Dara A. (@daradoescode) on X

    x.comSaved Oct 17#3

    Dara A. claims a small addition to her agents.md doubled the amount she ships, with her bio noting past work at Amazon and current agent-development projects.

    Why it matters. AGENTS.md is the convention your own agent harnesses live by; a claimed 2x shipping lift from a prompt-level change is worth testing yourself before treating it as gospel, but the experiment is cheap and directly transferable to your factory agents.

  4. McBain (@heyneighbor) on X

    x.comSaved Oct 16#4

    An engineer who previously worked at Lockheed Martin automated his Ocala pest-control business with OpenClaw, then turned it into small-business consulting; clients own their setup and data with no lock-in, and a 66-year-old admin uses OpenClaw daily.

    Why it matters. OpenClaw is the platform your software factory is built on, and this is a grounded, non-hype proof point of it automating a real small business end to end — useful both as a reference and as a case for how low-friction the tool can be for non-technical users.

02

Systems, craft, and tech debt

1 bookmark

  1. The Decoherence Trap: When AI slop drowns out the signal in an endless sea of noise

    maevyn.codesSaved Oct 17#5

    "Decoherence" is the buildup of locally valid but mutually incompatible models within a system, compounding with the rate of change; AI now produces it faster than ever, risking a "Decoherence Collapse." Patchwork, as-needed replacement is a trap that makes future changes harder to reason about.

    Why it matters. you evaluate agent-built systems on maintainability and you're explicit that LLMs shouldn't replace deterministic judgment in architecture; this essay is the strongest articulation of the failure mode you're designing your factory's review and consistency layers against.

03

Product infrastructure

1 bookmark

  1. New domain reputation and email warm-up - Email delivery | Clerk Docs

    clerk.comSaved Oct 16#6

    A new domain starts with zero reputation and may be throttled regardless of IP cleanliness; providers weigh domain age, SPF/DKIM/DMARC authentication, engagement, bounce rate, content, and sending patterns, each with their own requirements.

    Why it matters. you're shipping products like Clever Cat Prep that depend on deliverability from day one; this is practical reference material for getting a new domain's email reputation warmed up before launch traffic hits.