3314112bb9
Add two new sources with summaries, new concepts (developer-as-agent-manager, review-is-the-new-bottleneck), new entities (SWEPR, Nikolai Sheiko), and a query on the Stanford source; update related concept pages, overview, index, and log.
271 lines
62 KiB
Markdown
271 lines
62 KiB
Markdown
# Operation Log
|
||
|
||
Append-only chronological record of wiki operations. Newest entries at the bottom.
|
||
|
||
Entry format:
|
||
|
||
```
|
||
## YYYY-MM-DD — <operation>
|
||
- Intent: ingest | query | lint | maintenance
|
||
- Input: <source path / question / scope>
|
||
- Pages changed: [[...]], [[...]]
|
||
- Notes: <what changed, uncertainty, next steps>
|
||
```
|
||
|
||
---
|
||
|
||
## 2026-07-14 — maintenance
|
||
- Intent: maintenance
|
||
- Input: Initialize vault folder structure.
|
||
- Pages changed: created `index.md`, `log.md`, `wiki/overview.md`; created `raw/{sources,assets}` and `wiki/{sources,entities,concepts,timelines,comparisons,queries,lint-reports}`.
|
||
- Notes: Empty vault scaffolded per the LLM Wiki Schema in `CLAUDE.md`. No sources ingested yet. Ready for first "ingest <source>".
|
||
|
||
## 2026-07-14 — ingest (batch, 4 sources)
|
||
- Intent: ingest
|
||
- Input: The 4 talk/interview docs (user-selected scope; "Ideas for webinar" + "HR Contacts" left as raw reference).
|
||
- `raw/sources/Everything we knew about software has changed.md`
|
||
- `raw/sources/In 1 Year, the Gap Between AI Users and Everyone Else Will Be Irreversible.md`
|
||
- `raw/sources/sebastian interview - conclusions and insights.md`
|
||
- `raw/sources/Скиллы на базе git — новая память AI-агентов.md`
|
||
- Pages created — sources (4): [[2026-07-14-everything-we-knew-about-software-has-changed]], [[2026-07-14-gap-between-ai-users-irreversible]], [[2026-07-14-sebastian-eugene-interview]], [[2026-07-14-skills-based-on-git]].
|
||
- Pages created — entities (8): [[theo-browne]], [[allie-miller]], [[sebastian]], [[eugene]], [[konstantin]], [[claude-code]], [[hermes]], [[virtido]].
|
||
- Pages created — concepts (13): [[harness]], [[skills-as-memory]], [[evolution-of-agent-tooling]], [[agentic-loops]], [[context-as-scarce-resource]], [[personal-ai-operating-system]], [[product-ownership]], [[connections-as-moat]], [[seniority-and-the-junior-squeeze]], [[decoupling-identity-from-profession]], [[think-wider-not-bigger]], [[code-as-throwaway]], [[enterprise-ai-reality]].
|
||
- Pages created — timelines (1): [[ai-agent-evolution]].
|
||
- Pages updated: [[overview]] (added the cross-source through-line + agree/diverge), `index.md` (full catalog).
|
||
- Notes: Strong cross-source convergence (skills-as-memory: Konstantin ↔ Allie; connections: Sebastian ↔ Allie; code-as-throwaway: Theo ↔ Sebastian). Recorded contradictions (Eugene vs Sebastian on BYO- vs company-managed harness, online vs in-person networking, OSS motivation). Tentative claims flagged: Theo's model-era names, Eugene = vault owner, Hermes/Ralph heuristics, Allie's self-learning-models prediction.
|
||
- Next: consider ingesting "Ideas for webinar" (connects corpus → webinar plan) and "HR Contacts" (the HR-search demo); consider a `wiki/comparisons/` page for the BYO-vs-managed-harness debate; run "lint wiki" to check link health.
|
||
|
||
## 2026-07-14 — maintenance (tagging convention)
|
||
- Intent: maintenance
|
||
- Input: user request — add per-type hashtags to the wiki schema and existing pages.
|
||
- Pages changed: `CLAUDE.md` (new "Tagging Rules" section, rule 9, templates + lint checklist updated); all 26 wiki pages received a page-type tag line under their H1 — [[overview]] `#overview`, 4 source pages `#source`, 8 entity pages `#entity`, 13 concept pages `#concept`, [[ai-agent-evolution]] `#timeline`; `index.md` (noted the convention).
|
||
- Notes: Tag mapping is folder-driven (`#source`, `#entity`, `#concept`, `#timeline`, `#comparison`, `#query`, `#lint-report`, `#overview`). Exactly one type tag per page, first on the tag line; topical tags may follow. Lint now checks for missing/mismatched type tags.
|
||
|
||
## 2026-07-14 -- query
|
||
- Intent: query
|
||
- Input: "what would be the best claude skill to give to a claude beginner user?"
|
||
- Pages changed: created [[2026-07-14-best-first-skill-for-beginner]]; updated [[skills-as-memory]] (Next Questions item answered + linked), `index.md` (Queries section).
|
||
- Notes: Answer synthesized from Allie Miller (skill-creator, tone-of-voice/anti-AI-language starter set, foundation-docs-first) and Konstantin (>5-tool-calls auto-creation heuristic). Follow-up: contents of a good tone-of-voice SKILL.md; foundation-doc templates.
|
||
|
||
## 2026-07-14 — maintenance (authorized raw edit)
|
||
- Intent: maintenance
|
||
- Input: user explicitly authorized editing `raw/sources/Webinar script.md` this session (exception to rule 1).
|
||
- Pages changed: `raw/sources/Webinar script.md` — completed the dangling Tools→Memory bridge line, wrote the full **# Memory** section (amnesia demo → notebook tool → close/reopen payoff → system-prompt reveal → user.md beat → tiny-notebook limitation as the Skills hook), and renamed "Markus" → "Marcus" to match the demo app's seeded mail (`Marcus Hale`).
|
||
- Notes: Memory section is grounded in the actual mini-game behavior (Level 04 in `D:\Projects\Names\Webinars\From_chat_to_os`): frozen snapshot per session (hence the mandatory close/reopen), `memory.md` + `user.md` on disk, ~2200-char budget motivating Skills. No wiki pages changed. Next: draft **# Skills** section (open with on-demand loading answering the phone-book limitation).
|
||
|
||
## 2026-07-14 — maintenance (authorized raw edit, Skills section)
|
||
- Intent: maintenance
|
||
- Input: user request — draft the **# Skills** section in `raw/sources/Webinar script.md` (edit permission granted this session). Corrected flow per user: presenter dictates the operations step by step first, *then* asks AI to package the procedure as a skill.
|
||
- Pages changed: `raw/sources/Webinar script.md` — full Skills section (step-by-step dictation → "I was the recipe" → save-as-skill → SKILL.md on disk → fresh-session one-liner → progressive-disclosure reveal → portability beat → "I'm still the alarm clock" hook into Process).
|
||
- Notes: Grounded in the app's Skills level (Level 05): `save_skill`/`load_skill`, real `skills/<name>/SKILL.md`, live menu (names+descriptions only) rebuilt per window. No wiki pages changed. Next: draft **# Process** (Level 06 — spawn_process scheduler answers the alarm-clock limitation).
|
||
|
||
## 2026-07-14 — maintenance (authorized raw edit, Process section)
|
||
- Intent: maintenance
|
||
- Input: user request — draft the **# Process** section in `raw/sources/Webinar script.md` (session edit permission).
|
||
- Pages changed: `raw/sources/Webinar script.md` — full Process section (goal not task → spawn_process with pid → hands-off heartbeats → Override slider makes the cube move by itself → ps/kill → reveal: the shell authored the worker's prompt (AI managing AI) → closing arc: same model all six levels, only the harness grew; harness is unique to you; "from a chat box to your own Agentic Operating System"). Also two-branch rule fixes earlier this session (Memory convention + Skills dictation).
|
||
- Notes: Grounded in Level 06 design (`2026-07-11-process-level-design.md`): single ~5s scheduler, stateless ticks, spawn/list/stop syscalls, live −5…+5 weather offset. The "same model, growing harness" close absorbs the `_ideas:` note at the top of the script. Script now has all sections drafted (Intro → Process).
|
||
|
||
## 2026-07-14 — maintenance (authorized raw edit, OS section)
|
||
- Intent: maintenance
|
||
- Input: user request — add an **# OS** section to `raw/sources/Webinar script.md` (session edit permission): typing removed, mouse-only interaction with the agent.
|
||
- Pages changed: `raw/sources/Webinar script.md` — Process now ends with a "still a terminal / most people never will type" hook; new OS section (one-button app → rule baked in → 5-second checkbox → appliance → "the agent became a program / the chat box dissolved into the OS"); the "same model, growing harness" closing arc moved from Process into OS, with the new final callback "a button that already knows what the email said."
|
||
- Notes: Grounded in the OS app design (`2026-07-12-os-app-design.md`): tile 07, status bar mirrors the agent's closing sentence, shared scheduler slot, Override interplay, rule baked into the prompt. Script ladder now: Chat box → ReAct → Tools → Memory → Skills → Process → OS.
|
||
|
||
## 2026-07-14 — maintenance (authorized raw edit, OS reframing)
|
||
- Intent: maintenance
|
||
- Input: user correction — the webinar's core idea is "make tools for yourself," not "become a product."
|
||
- Pages changed: `raw/sources/Webinar script.md` — Process→OS hook now ends "…and become a tool? A small one. Made for exactly one person."; OS landing gained the intro callback (weekends burned on internal tools → this one took an evening, asked-for not written) and "It dissolved — into the operating system. Into little tools you make for yourself."; closing arc adds "You don't buy it. You build it — one small tool at a time."
|
||
- Notes: The OS section now closes the loop with the Intro's internal-tools passion. No wiki pages changed.
|
||
|
||
## 2026-07-14 — ingest (batch, 2 sources: Nina + Yulia interviews)
|
||
- Intent: ingest
|
||
- Input: `raw/sources/Nina interview.md`, `raw/sources/Yulia interview.md` (both are conclusions docs auto-generated by Eugene's record→diarize→transcribe→summarize tool).
|
||
- Pages created — sources (2): [[2026-07-14-nina-interview]], [[2026-07-14-yulia-interview]].
|
||
- Pages created — entities (3): [[nina]] (HR recruiter, Virtido), [[yulia]] (HR lead; name/affiliation tentative, inferred from filename + Nina reference), [[inspectron]] (Eugene's employer).
|
||
- Pages created — concepts (2): [[levels-of-ai-usage]] (chatbot → memory → Claude Code → CLAUDE.md → skills → KB → RAG; non-programmer ceiling = CLAUDE.md + skills), [[solve-first-then-skillify]] (solve once, then freeze into a skill; ~3-message and >5-tool-call heuristics).
|
||
- Pages updated: [[eugene]] (identity evidence strengthened — webinar author/presenter, Inspectron, $200 max plan, tool builder; still tentative), [[virtido]] (HR team = webinar audience, pain points, tooling), [[skills-as-memory]] (handoff/de-risking payoff, ~3-message heuristic, method link), [[connections-as-moat]] (Eugene independently converges: "the human's role is just to be human"), [[overview]] (6 sources, demand-side bullet, agree/diverge, navigation, open questions), `index.md`.
|
||
- Notes: Strong convergence — Eugene ↔ Sebastian/Allie on human-connection residual despite their networking-tactics disagreement; Eugene's ~3-message heuristic ↔ Konstantin's >5-tool-call heuristic. New cross-interview finding: adoption blocked by friction, not resistance. Tentative flags: Yulia's name/affiliation; Eugene's price-rise prediction; Inspectron employment-vs-contracting. Skills-IP-vs-work-product dispute (Eugene vs Sebastian) recorded, unresolved.
|
||
- Next: ingest "Ideas for webinar" / "Webinar Plan" / "Webinar script" to connect corpus → deliverable; interview Larysa/Larisa (PM use cases, twice flagged); consider a comparison page "personal vs employer ownership of skills".
|
||
|
||
## 2026-07-14 — query (network from a standing start)
|
||
- Intent: query
|
||
- Input: "how does an individual actually build a network from a standing start?" + "go" on the recommended plan (tentative protocol + Sebastian follow-up instrument).
|
||
- Pages created: [[network-from-a-standing-start]] (concept, Status: tentative — 5-step protocol: recurring venues → cadence → lead with humanness → engineer second meetings → track second meetings/let referrals work; contested anti-tactics list; webinar flagged as live case study), [[2026-07-14-network-from-standing-start]] (query — 10-question Sebastian round-2 interview instrument in 3 blocks: bootstrap biography / mechanics / falsification of "Big zero").
|
||
- Pages updated: [[connections-as-moat]] (open question → protocol + validation plan), [[overview]] (vault-level open question annotated), `index.md` (Concepts + Queries sections).
|
||
- Notes: Corpus cannot answer the question directly — protocol is inferred from Sebastian's principles, Allie's prediction, and Eugene's convergent thesis; every claim marked tentative/contested. Key falsification target: whether Sebastian ever had a true standing start (prior-job network as seed). Deep-research complement (weak ties, mere exposure, from-zero playbooks, scoped to test "Big zero") described but not run.
|
||
- Next: run the round-2 Sebastian interview and ingest it; decide on the deep-research run; track webinar → paid-HR-build chain as case-study evidence.
|
||
|
||
## 2026-07-16 — query → comparison
|
||
- Intent: query
|
||
- Input: "Compare Theo's, Konstantin's ('Sonstantine') and Allie's approaches — what they have in common, what they have different?"
|
||
- Pages created: [[theo-konstantin-allie]] (first `wiki/comparisons/` page — three-lens side-by-side: at-a-glance table, 6 shared points, differences by altitude/framing/code/learning/register, complementary tensions).
|
||
- Pages updated: [[theo-browne]], [[konstantin]], [[allie-miller]] (added inbound "Comparison:" links to avoid orphan); `index.md` (Comparisons section — first entry).
|
||
- Notes: Synthesized from the 3 source summaries + entity pages + [[skills-as-memory]]. Common ground: bottleneck moved model→human system; markdown/skills as the atomic unit; infrastructure compounds; system-ready to absorb new models; context as scarce resource; shared [[claude-code]]. Differences framed as *altitude* (Theo strategy/psychology · Konstantin engineering · Allie individual-productivity), not direction — corpus disagreements lie elsewhere. Recorded the throwaway-code (Theo) vs persist-in-git (Konstantin) tension as complementary (artifact vs capability). No new contradictions. Saved as a comparison rather than a query page since the request was explicitly a side-by-side.
|
||
- Next: reusable starter template (foundation docs / good SKILL.md) still unresolved across all three — webinar-relevant; consider whether Konstantin's meta-loop (wipe-and-restart) tensions with "persist everything".
|
||
|
||
## 2026-07-16 — maintenance (table fix)
|
||
- Intent: maintenance
|
||
- Input: user report — the table in [[ai-agent-evolution]] "looks broken".
|
||
- Pages changed: [[ai-agent-evolution]] — repaired the "Agent architecture & tooling (Konstantin)" table; escaped the `|` inside aliased wiki links (`[[harness\|harnesses]]`, `[[skills-as-memory\|skills]]`, `[[agentic-loops\|agent loops]]`) so Obsidian no longer reads them as column separators, and removed the two phantom trailing columns the unescaped pipes had forced into the header/separator. Now a clean 3-column table (Period / Milestone / Significance).
|
||
- Notes: Content unchanged — formatting-only fix. The Theo Browne table was already valid. No `index.md` change (no catalog/structural change). Root cause: piped `[[page|alias]]` links inside GFM/Obsidian table cells require the pipe escaped as `\|`.
|
||
|
||
## 2026-07-21 — ingest (Larysa interview)
|
||
- Intent: ingest
|
||
- Input: "ingest Larysa interview.md" → `raw/sources/Larysa interview.md` (auto-generated interview conclusions doc; Eugene × Larysa, technical BA/PM and ex-mobile dev).
|
||
- Pages created: [[2026-07-21-larysa-interview]] (source), [[larysa]] (entity), [[integration-dead-ends]] (concept — agent begins work against connectors gated by account tier / paid seat / missing API; corpus offers no fix, only up-front verification), [[leave-less-room-for-imagination]] (concept — under-specified prompts drift and the collateral damage is *invisible*; skills as the constraint; includes the 4.7-over-4.8 model-choice corollary).
|
||
- Pages updated: [[skills-as-memory]] (new "negative case" section — built-in memory as anti-feature, the demand-side reason skills exist; contradiction logged vs Allie), [[harness]] (consolidation-over-tool-hopping section + "before and after" claim marked self-reported), [[code-as-throwaway]] (auth/payments trust carve-out; "safe from 4.6 on"), [[levels-of-ai-usage]] (Larysa as proof the rungs are skippable — technically advanced, architecturally stuck), [[claude-code]] (practitioner-limits section: memory, connectors, 4.6/4.7/4.8, emulator), [[solve-first-then-skillify]], [[personal-ai-operating-system]] (cross-ref to Eugene's convergent OS framing), [[eugene]], [[yulia]] (Larisa open question partially resolved), [[virtido]] (audience extends beyond HR; ClickUp/Figma/Teams/Slack), [[overview]] (6→7 sources, demand-side paragraph, two new divergences, two new vault-level open questions), `index.md`.
|
||
- Notes: Closes the "Larysa/Larisa not yet interviewed — PM use cases missing" gap flagged in both HR interviews. Key reframe for the webinar: this source moves the diagnosis from *friction* (Nina/Yulia: people would adopt if it were simple) to *structural walls* for users already past friction — memory, entitlements, drift. Two genuine corpus contradictions recorded rather than smoothed: memory-as-anti-feature (Eugene) vs Allie's untroubled use of persistent context docs, and tight-spec vs [[think-wider-not-bigger]]. Also corrected `index.md`: `HR Contacts.md` was listed as not-yet-ingested but does not exist in `raw/sources/` — removed from the list.
|
||
- Next: remaining un-ingested raw are the three webinar-deliverable docs (`Ideas for webinar.md`, `Webinar Plan - From Chat Box to Your Own OS.md`, `Webinar script.md`) — ingesting them would connect the corpus to the actual deliverable and settle whether the script keeps the [[levels-of-ai-usage]] rung order. Candidate query: does the skills rung actually answer Larysa's memory complaint, or is hers a cross-project problem skills don't solve? A first lint pass is also overdue (no lint reports exist; 7 sources, 17 concepts).
|
||
|
||
## 2026-07-22 — query (webinar theses)
|
||
- Intent: query
|
||
- Input: "What theses can I suggest for the webinar ('from chatbox to your own agentic operating system') based on what you already have?"
|
||
- Pages changed: created [[2026-07-22-webinar-theses]]; updated `index.md` (Queries section).
|
||
- Notes: 14 candidate theses synthesized from [[overview]], the 7 source summaries, and direct reads of the three raw webinar-deliverable docs (Plan, script, Ideas — still raw-only, cited as raw per policy). Grouped: spine (harness-not-model, app-you-open→OS, skills-as-memory, context scarcity, build-not-buy) / stakes (cost→zero, irreversible gap, connections rise) / obstacles (friction-not-resistance, structural walls, less-room-for-imagination) / method (solve-first-then-skillify, research-vs-judgment, walk-in-a-week) / honest tensions (3 recorded contradictions offered as Q&A ammo). Recommended 7-thesis subset for the 30-min format noted in the query page. No concept/entity pages changed.
|
||
- Next: ingest the three deliverable docs so theses can cite wiki sources instead of raw; decide whether the "obstacles" theses earn a station on the spine.
|
||
|
||
## 2026-07-22 — ingest (ИИ глупый! short)
|
||
- Intent: ingest
|
||
- Input: "ingest \"ИИ глупый!\"" → `raw/sources/ИИ глупый!.md` (conclusions doc for a 1:28 Russian YouTube Short, author unknown; file was added mid-session after the first lookup found nothing).
|
||
- Pages created: [[2026-07-22-ai-is-stupid]] (source).
|
||
- Pages updated: [[harness]] (business-facing "engineering wrapper" definition + model+context+harness formula, flagged as webinar language), [[context-as-scarce-resource]] (supply-side facet: "intelligence without context loses to context without intelligence", Nobel-vs-employee analogy; RAG-vs-skills tension logged under Contradictions), [[overview]] (7→8 sources, convergence one-liner, navigation), `index.md` (Sources entry; `my theses.md` added to the not-yet-ingested list).
|
||
- Notes: No new concepts — the short *restates* the vault's machine-side spine for a business audience, which makes it webinar-quotable rather than novel. One tension recorded, not smoothed: it recommends RAG/long-term memory as context mechanisms where Konstantin argues skills beat RAG (possibly audience-driven — business data vs procedures). Authorship unknown and marked tentative; its framing matches the corpus closely enough that a connection to someone already in the corpus is worth checking. All ×10…×1000 gain claims are rhetorical, not measured.
|
||
- Next: `my theses.md` is new in raw and un-ingested (six webinar thesis one-liners) — small, high-relevance ingest candidate; the three webinar-deliverable docs remain pending; first lint pass still overdue (8 sources, 17 concepts, 0 lint reports).
|
||
|
||
## 2026-07-24 — ingest (You're reading way too much code)
|
||
- Intent: ingest
|
||
- Input: "ingest \"You're reading way too much code\"" → `raw/sources/You're reading way too much code.md` (conclusions doc for Theo Browne's 24:11 video; his second source in the vault).
|
||
- Pages created: [[2026-07-24-youre-reading-way-too-much-code]] (source), [[make-more-cheap-code]] (concept — ship/no-ship line, four tiers of code A–D, 100–10,000:1 slop-to-ship verification ratio, always-another-layer, dumb-model agents as API usability testers, reading-costs-attention economics).
|
||
- Pages updated: [[code-as-throwaway]] (discipline section; evidence; its "what is the durable artifact" open question partially answered — the verification harness is now a named first-class output), [[theo-browne]] (second talk, recurring author-move noted, self-reported ratios flagged tentative), [[leave-less-room-for-imagination]] (new logged tension: Theo/Dax agent-diff-summaries vs Eugene's invisible-drift claim), [[2026-07-14-everything-we-knew-about-software-has-changed]] (reciprocal related-source link), [[overview]] (8→9 sources; verifying-half of spine given its method; new divergence line; 17→18 concepts), `index.md` (Sources, People, Concepts entries).
|
||
- Notes: The source is a *defense with discipline*, not a reversal — Theo explicitly rejects shipping unreviewed slop, which pre-empts the obvious objection to [[code-as-throwaway]]. Strong webinar relevance: "attention, not generation, is the bottleneck" is the engineer-side twin of [[context-as-scarce-resource]], and the tier framework gives audiences a non-binary answer to "can I trust AI code?". One genuine tension recorded rather than smoothed (summaries-hide-drift). Shao's 80%-into-harnesses ratio and all of Theo's daily-line counts are self-reported/uncited — marked tentative. Open framework gap flagged: skills/CLAUDE.md files don't fit the A–D spectrum.
|
||
- Next: candidate query — what is the non-engineer's analog of throwaway verification code (HR/BA audience)? `my theses.md` and the three webinar-deliverable docs remain un-ingested; first lint pass still overdue (9 sources, 18 concepts, 0 lint reports).
|
||
|
||
## 2026-07-24 — query (non-engineer throwaway verification)
|
||
- Intent: query
|
||
- Input: "query: what is the non-engineer's analog of throwaway verification code?" (the candidate query flagged in the previous ingest).
|
||
- Pages created: [[2026-07-24-non-engineer-throwaway-verification]] (query — answer: disposable AI work that attacks/misreads/simulates the deliverable before a human sees it; one-to-one mapping table from Theo's engineer patterns to HR/BA use cases; "checker skills" as the second species of skill; tier-D-stays-human caveat since non-engineer verification bottoms out in judgment, not tests).
|
||
- Pages updated: [[make-more-cheap-code]] (next-question marked answered, link added), [[leave-less-room-for-imagination]] (new next-question: drift harnessed as a sandboxed ambiguity diagnostic — synthesis, tentative), `index.md` (Queries section).
|
||
- Notes: Pure synthesis, no new source — every mapped pattern is grounded in corpus use cases (Nina/Yulia job descriptions, candidate KB, sourcing; Larysa spec ambiguity; Allie's anti-AI-language checker-skill precedent). Key reframes with durable value: (1) the fresh-agent misread test inverts [[leave-less-room-for-imagination]] — drift becomes a diagnostic when sandboxed; (2) [[solve-first-then-skillify]] populates skills in two species, producers and checkers. Flagged honestly: the whole mapping is argument-by-analogy with no measured claim; Nina's transcript-beats-summary finding is a standing counterweight to summary-based review.
|
||
- Next: decide whether "checker skills" earns a slide on the webinar's skills rung; the webinar-deliverable docs (`Ideas`, `Plan`, `script`, `my theses.md`) remain un-ingested; first lint pass still overdue (9 sources, 18 concepts, 2026-07-24 now has 1 query, 0 lint reports).
|
||
|
||
## 2026-07-28 — ingest (Agentic Engineering, explained by a 10x developer)
|
||
- Intent: ingest
|
||
- Input: "ingest \"Agentic Engineering, explained by a 10x developer\"" → `raw/sources/Agentic Engineering, explained by a 10x developer.md` (conclusions doc for a 42:33 YouTube interview; Thorsten Ball × David Andre). The densest single source in the vault so far.
|
||
- Pages created — source (1): [[2026-07-28-agentic-engineering-10x-developer]].
|
||
- Pages created — entities (2): [[thorsten-ball]] (founding engineer at AMP; author of the Go interpreter/compiler books), [[amp]] (Sourcegraph's agent — the vault's second reference harness).
|
||
- Pages created — concepts (5): [[shedding-weight]] (delete process that only existed because humans were the bottleneck — backlogs, CI that repeats the agent's tests, IDE extensions, local dev), [[build-for-the-agent-not-the-human]] (no forms; the admin panel that dies; bring your own agent), [[emacsification-of-software]] (fork-and-remix, never upstream; software becomes bespoke), [[explosion-of-internal-software]] (the Excel/wiki/hack layer becomes real tools; skill + token budget as the two dividing variables), [[async-by-default]] (orbs as remote sandboxes; one URL = thread + agent + computation + diff; ask for proof since you're waiting anyway).
|
||
- Pages updated — concepts (10): [[skills-as-memory]] (**the dissent**, plus three competing readings under Contradictions), [[evolution-of-agent-tooling]] ("a fourth position: skip the progression"; A2A partially answered by AMP's agent-to-agent messaging), [[harness]] (AMP as second reference harness; vendor-managed as a third governance option; local-vs-remote tension), [[context-as-scarce-resource]] (information > tuning; the two information sources; context as scarce per-*wallet*), [[code-as-throwaway]] (99%-AI-written production datapoint; slop-is-human), [[make-more-cheap-code]] (variations-not-answers; ask-for-proof; attention-vs-token-budget sharpening), [[product-ownership]] (first-principles as the top skill; the printer/tablet push-back; what got commoditised), [[seniority-and-the-junior-squeeze]] (2–3 years of hand-taught knowledge → 30-second output; "don't compare yourself to the 1%"), [[enterprise-ai-reality]] (token budget as a second divide; why the frontier playbook doesn't transfer), [[leave-less-room-for-imagination]] (his 5-part prompt structure as a worked example; model-choice tension), [[personal-ai-operating-system]] (the fourth layer: tools you build for yourself).
|
||
- Pages updated — entities (2): [[theo-browne]] (ally cross-link), [[claude-code]] (AMP as the contrasting harness design). Plus [[overview]] (9→10 sources; new "frontier side" of the through-line; agree/diverge rewritten; 3 new vault-level open questions) and `index.md`.
|
||
- Notes: **The important thing this source does is contradict the vault's spine.** Thorsten ships a 99%-AI-written codebase with no skills, no MCP servers and no slash commands — his context lives in the codebase and `AGENTS.md`. This is the first credible rejection of the mechanism the webinar's central promise rests on, and the evidential asymmetry favours him (first-hand daily practice at scale vs Konstantin's architecture argument plus self-reported individual workflows). Recorded as three competing readings on [[skills-as-memory]] — situational (he owns one codebase; the HR/BA audience owns none) / premature abstraction / same thing under another name (AMP's vendor-curated sub-agents and `AGENTS.md` *are* two-stage context, just not user-authored). Not smoothed, not resolved. Second-order tensions logged: model choice as lever (Eugene 4.7-over-4.8) vs distraction (Thorsten); local consolidated workspace (Eugene) vs local dev disappearing (Thorsten); shed-your-process (frontier) vs compliance-is-the-deliverable ([[enterprise-ai-reality]]).
|
||
- Also strongly *confirmatory*: the club food-ordering app (menu photo → working app, then ~2 hours of phone typing to encode a 20-person club's ordering process) is the corpus's best outside evidence for the webinar's "little tools you make for yourself" thesis — reached independently, by an engineer, applied to non-technical people. And "slop comes from humans" converges with Theo from a different direction (taste vs verification discipline).
|
||
- New named variable with no answer anywhere in the corpus: **token budget** as one of two winner/loser variables. Elevated to a vault-level open question.
|
||
- All quantities self-reported from inside the company selling the agent (99% figure, the team poll, velocity claims) — flagged tentative throughout. "Local dev is going away" is a prediction from a remote-sandbox vendor — flagged.
|
||
- Housekeeping: corrected the stale `index.md` "not yet ingested" list — those four webinar docs moved to `raw/notes/` (commit 1361dd7) and are authored deliverables, not sources. `raw/sources/` is now fully ingested (10/10).
|
||
- Next: the highest-value move is now a **lint pass** — 10 sources, 24 concepts, 0 lint reports, and this ingest added five concepts plus a live contradiction that touches the spine. Candidate query with real stakes for the webinar: *does the skills rung survive Thorsten's counter-example, and what would test it?* (the cheap experiment — same task with and without a skill, in a non-engineer's hands — has never been run). Also worth considering: a comparison page "Thorsten vs the skills camp", and whether [[shedding-weight]] / [[explosion-of-internal-software]] earn stations on the webinar spine.
|
||
|
||
## 2026-07-28 — lint (first pass)
|
||
- Intent: lint
|
||
- Input: "lint wiki" — first health check on the vault (10 sources · 24 concepts · 14 entities · 4 queries · 1 comparison · 1 timeline; 0 prior lint reports).
|
||
- Pages created: [[2026-07-28-lint]] (first `wiki/lint-reports/` page).
|
||
- Checks run: page-type tags vs folder + line-3 placement (41/41 pass) · H1 presence (41/41) · required template sections for sources (10/10) and entity/concept pages (38/38) · Evidence sections citing ≥1 source page (38/38) · broken `[[links]]` (0) · orphans (0) · zero-outbound pages (0) · `index.md` ↔ `wiki/concepts/` drift (0, 24/24 in sync) · `raw/` path resolution (5 failures) · contradiction reciprocity · high-mention concepts lacking pages · link-graph inbound/outbound counts.
|
||
- Headline: **structurally clean; the real defects were all staleness of synthesis.** 12 findings, 9 fixed in this pass, 3 left as recommendations because they are scope decisions rather than defects.
|
||
- Fixes applied — references (3): stale `raw/sources/` paths for the four deliverable docs that moved to `raw/notes/` in commit 1361dd7, corrected in [[levels-of-ai-usage]], [[2026-07-22-webinar-theses]] and [[2026-07-14-sebastian-eugene-interview]] (5 refs, reworded "not yet ingested" → "authored deliverable"); [[claude-code]] summary said "cited across all four ingested sources" (written at 4 sources, now 10) — rewritten to a claim that won't rot, noting Thorsten as the sole practitioner on a different harness; [[overview]] open question naming `HR Contacts` (a file that does not exist — corrected in `index.md` on 2026-07-21 but never propagated here) marked resolved.
|
||
- Fixes applied — one-sided contradictions (4): rule 5 requires contradictions be recorded explicitly, and four were logged on one page but not on the page holding the opposing view. Added reciprocal entries to [[think-wider-not-bigger]] (vs tight specs), [[levels-of-ai-usage]] (the skills dissent — its *top rung* is what Thorsten contests), [[personal-ai-operating-system]] (memory-as-anti-feature vs its *layer 1*, plus the skills dissent vs its layer 2), [[skills-as-memory]] (skills-vs-RAG, which its own Summary asserts as settled). L5 and L6 were the consequential ones: in both, the page *making* the contested claim was the page not carrying the objection.
|
||
- Fixes applied — staleness notes (2): [[2026-07-22-webinar-theses]] (synthesized from 7 sources, 10 now exist; thesis 3 "Skills are the new memory" is load-bearing and its counter-example was missing from its own tensions list; nothing from the last two ingests appears) and [[theo-konstantin-allie]] (predates Theo's second source and Thorsten; its closing "none of the three directly contradicts another" now misleads about the skills thread). Both preserved rather than rewritten, per the Update Policy's no-silent-large-rewrites rule.
|
||
- Open findings (not fixed — scope decisions): **(1)** "taste" is the vault's largest uncovered concept — 12 pages / 17 occurrences, one of four nouns in the through-line, the other three all have pages; named the meta-skill by Allie and the whole slop answer by Thorsten; recommend a concept page. **(2)** Query pages are a weakly-linked class (0/1/2/2 inbound); [[2026-07-22-webinar-theses]] is effectively orphaned despite being the most webinar-relevant page in the vault. **(3)** Watch items: [[2026-07-22-ai-is-stupid]] is thinly integrated (2 citing pages vs a median of ~17, plausibly correct since it restates rather than adds); [[emacsification-of-software]] + [[explosion-of-internal-software]] should merge if neither gains second-source support by the next lint. Considered and rejected as a new page: "trust calibration" — covered by [[make-more-cheap-code]] + [[code-as-throwaway]] + [[seniority-and-the-junior-squeeze]]; creating it would duplicate rather than consolidate.
|
||
- Contradiction inventory: 9 live disagreements, now all recorded on both sides. #1 (skills-as-memory vs no-skills-at-all) is the vault's most consequential open question because the webinar's central promise rests on it.
|
||
- Next: refresh [[2026-07-22-webinar-theses]] against all 10 sources (fixes the staleness and most of the weak-linking in one operation, and is the page closest to the deliverable); create the `taste` concept page; extend [[theo-konstantin-allie]] to four lenses. The falsification test flagged on [[skills-as-memory]] — same task with and without a skill, in a non-engineer's hands — remains the cheapest experiment that would move contradiction #1.
|
||
|
||
## 2026-07-28 — query (webinar theses v2 — refresh)
|
||
- Intent: query
|
||
- Input: "Refresh 2026-07-22-webinar-theses" — acting on the top recommendation from [[2026-07-28-lint]] (v1 was two ingests stale and effectively orphaned).
|
||
- Pages created: [[2026-07-28-webinar-theses]] (17 theses from all 10 sources + the current deliverable state in `raw/notes/`).
|
||
- Pages updated: [[2026-07-22-webinar-theses]] (staleness note replaced with a **SUPERSEDED** banner summarising what changed and pointing forward — preserved as the 7-source state, not overwritten, per the Update Policy); backlinks added from [[explosion-of-internal-software]], [[skills-as-memory]], [[shedding-weight]] and [[2026-07-24-non-engineer-throwaway-verification]] so v2 does not repeat v1's orphaning; `index.md` (Queries section marks v1 superseded, v2 current).
|
||
- Method note: created as a **new dated page** rather than editing v1 in place. Query pages are dated Q&A snapshots (`wiki/queries/YYYY-MM-DD-<slug>.md`); rewriting the 07-22 file would have made its date lie and would have been a silent large rewrite of a dated artifact.
|
||
- Substantive changes to the thesis set: **(1) T3 reframed** from "skills are the new memory" to *"context you author beats context that's inferred"* — the v1 wording has a live counter-example in [[thorsten-ball]], and the reframe is what all four practitioners actually agree on (Konstantin's skills, Allie's foundation docs, Eugene's anti-memory position, Thorsten's `AGENTS.md` are all authored context). It survives all three readings logged on [[skills-as-memory]] and keeps the script's Memory→Skills stations intact. **(2) T5 upgraded from assertion to evidence** — [[explosion-of-internal-software]] supplies a non-engineer-shaped outside case (20-person club, phone, menu photo, ~2 hours), which defuses the "sure, *you* can do that, you're technical" objection against the talk's least-provable claim. **(3) Three new theses** the earlier set had no source for: T6a verification/ask-for-checks, T10 shedding weight ("which of your processes only exist because *you* were the bottleneck?"), T13a ask-for-15-options. **(4) T1 strengthened** — Thorsten's "stop tuning model choice; the dominant variable is the information you put in" makes harness-not-model the best-evidenced claim in the vault, and it is already the script's literal closing argument. **(5) New honest caveat** — token budget as a second, non-skill axis of the gap.
|
||
- Revised 30-min cut: T2 · T4 · T3 · T6a · T1 · T5 · T15, mapped to script beats. Changed from v1's recommendation: T3 reframed, T6a added, T12 promoted (it now has a teachable five-part prompt structure from Thorsten rather than an abstract principle), T9 demoted to Q&A.
|
||
- Also produced: a Q&A-ammo section stating the skills dissent honestly with the three readings and the presentation-safe framing, and **three gaps the refresh exposes in the current script** — no verification beat (the audience's first question is "can I trust it?" and the script never once shows the agent being checked), no "what stays yours" beat (the closing arc is entirely harness, not human), and the unsaid token-budget caveat against the "a laptop, one hour, and your real work" promise.
|
||
- Read directly for this refresh: `raw/notes/Webinar script.md` (current ladder: Chat box → ReAct → Tools → Memory → Skills → Process → OS), `raw/notes/Webinar Plan - From Chat Box to Your Own OS.md`, `raw/notes/my theses.md`. Notable: T4 and T1 are already dramatized in the script better than a slide could do it ("the notebook is tiny. On purpose."; "the model never changed").
|
||
- Next: decide whether T6a earns a station or one line inside the Skills station (recommend the latter — a checker skill is one sentence of demo and costs no new level); decide whether T10 opens the talk rather than closing it; the remaining lint recommendations stand (create a `taste` concept page; extend [[theo-konstantin-allie]] to four lenses). The falsification test on [[skills-as-memory]] would settle T3's tension and would itself make a strong demo.
|
||
|
||
## 2026-07-28 — maintenance (token-budget claim rescoped; webinar gap #3 withdrawn)
|
||
- Intent: maintenance
|
||
- Input: user objection to gap #3 in [[2026-07-28-webinar-theses]] — every LM vendor sells a subscription, that subscription covers even advanced users, so "budget" reduces to subscription cost, which is obvious.
|
||
- Assessment: **objection upheld, and the corpus supports it more strongly than the original write-up did.** The token-budget claim was recorded at ingest as an unqualified "the divide is also a *spending* gap" without weighing three counter-datapoints already in the vault: [[eugene]] runs 7 project-agents in parallel on a $200 plan; [[allie-miller]] runs ~100 agents and 36 workflows; and [[2026-07-14-sebastian-eugene-interview]] frames levelling as "a 20-year veteran and a fresh grad **on the same subscription**." No practitioner in the corpus reports a cost ceiling. The claim's real scope is **metered** pricing — [[amp]] sells usage, and Thorsten's pattern is parallel remote sandboxes and parked orbs ([[async-by-default]]) — which is a fleet cost, not a seat cost.
|
||
- Pages changed: [[enterprise-ai-reality]] (scoping sub-bullet added under the token-budget item, with the counter-evidence named), [[explosion-of-internal-software]] (the "two variables" bullet rescoped; its Contradictions entry partly resolved and marked tentative), [[context-as-scarce-resource]] ("context now has a price" → "…at fleet scale"; per-request scarcity restated as the binding constraint for individuals), [[overview]] (vault-level open question narrowed to fleet/enterprise allocation), [[2026-07-28-webinar-theses]] (gap #3 withdrawn with the reasoning recorded inline; T7's "second axis" line rescoped so the thesis stays on the skill gap).
|
||
- Claim preserved, not deleted, per rule 5 — Thorsten did say it and it stands in its own regime. What changed is scope and the counter-evidence, both now stated on every page carrying it.
|
||
- Residual open: metered/fleet pricing and enterprise budget allocation remain unanswered; [[eugene]]'s price-rise prediction ("what I now buy for 200 will cost about 1,000") would reopen the question for individuals if it holds — currently a forecast, not a constraint. Status: tentative.
|
||
- Notes: worth flagging as a process lesson — the claim came from a credible source and was written up the same session it arrived, without checking it against the vault's existing practitioner evidence. A source's framing of its own economics is not automatically the corpus's.
|
||
|
||
## 2026-07-28 — query (verification beat design)
|
||
- Intent: query
|
||
- Input: "What are your suggestions for verification beat? What can we add?" — following the gap flagged in [[2026-07-28-webinar-theses]].
|
||
- Pages created: [[2026-07-28-verification-beat-design]] (placement analysis, three options costed by seconds, drafted script copy in the script's voice, audience-translation lines, honest caveats).
|
||
- Pages updated: [[2026-07-28-webinar-theses]] (T6a now points to the design page); `index.md` (Queries).
|
||
- Key synthesis: **the verification beat and the "what stays yours" beat are the same beat** — verification is exactly where the human's remaining job lives ([[product-ownership]]) — so one insertion closes both gaps the refresh identified. Matters for a 30-min format.
|
||
- Placement argument: the audience's unease peaks at one specific existing line in the Process station ("I take my hands off the keyboard… Nobody is typing. It just... runs."). Answer it there, not in Q&A. Recommended split: introduce the checker skill at **Skills** (~60–90s), cash it in at **Process** (~20s, reuses an existing reveal), put the judgment half in the **closing arc** (~30s).
|
||
- Three options by cost: **(1)** ~15s and free — the Process station already reveals the prompt the shell wrote for its worker; add one visible self-check line to that artifact, no new demo steps. **(2)** ~60–90s recommended — a second skill whose only job is to check the first, demonstrated by breaking the state *by hand* so the audience sees the error before the agent reports it; introduces the producer/checker species split from [[2026-07-24-non-engineer-throwaway-verification]] at no new level. **(3)** ~30s standalone — **the script already contains a perfect ambiguity example**: the Process goal line says "keep the cube on the **right shelf**", which reads as *correct* shelf or *right-hand* shelf. The colon disambiguates it as written, so the unsafe version can be shown deliberately as a live demo of [[leave-less-room-for-imagination]] (currently thesis T12 with no demo) and as motivation for the checker.
|
||
- Closing half drafted to fill gap #2: a checker catches drift but can never catch a wrong rule ("if twenty degrees was the wrong number, both agents agree, confidently, forever") → "the machine checks whether the thing was done right; you check whether it was the right thing." Converts the ending from capability to the audience's own value.
|
||
- Caveats recorded rather than smoothed: a checker authored by the agent it checks is not independent verification (the same limit already logged on [[async-by-default]] — proof produced by the thing being checked is evidence, not verification); [[nina]]'s transcript-beats-summary finding is a standing counterweight against implying reading is now optional; and none of this has been tested on a non-engineer audience, which is why the audience-translation lines after option 2 are load-bearing rather than optional.
|
||
- Note: `raw/notes/Webinar script.md` was read but **not modified** (rule 1). All drafted copy lives in the wiki for the user to apply.
|
||
- Next: gap #2 is now addressed by this design; the remaining lint recommendations stand (create a `taste` concept page; extend [[theo-konstantin-allie]] to four lenses). Open decision from the refresh — whether T10 ("which processes only exist because you were the bottleneck?") opens the talk.
|
||
|
||
## 2026-07-28 — maintenance (authorized raw edit: "right shelf" ambiguity demo) + theses legend
|
||
- Intent: maintenance
|
||
- Input: user — "That's a great idea about the 'right shelf'. would be a perfect demo. add it." Plus a question about the `T<number>` shorthand.
|
||
- **Authorized raw edit** (exception to rule 1, on direct instruction; same pattern as the 2026-07-14 script edits logged above): `raw/notes/Webinar script.md`, Process station. Inserted the ambiguity demo between "…but a goal." and the real goal line — deliberately sloppy prompt ("keep the cube on the right shelf") → agent moves the cube to the right-*hand* shelf → "That's not what I meant. I meant the *correct* shelf. It heard: the shelf on the *right*." → "It didn't ask. It didn't hesitate… It just confidently did the wrong thing." → "Every gap you leave, it fills. And it fills it silently." → ties into the hands-off moment a minute later and reuses Marcus's own word: "Marcus said: don't let it drift. Turns out the first thing that drifts… is what I meant." → then the precise goal as originally written.
|
||
- Placement rationale: kept at Process rather than moved to Skills (the design page's alternative) because the user is adding the ambiguity demo *alone*, without the checker skill — standalone it is strongest where the ambiguous phrase already lives and where handing over control is imminent. If the checker skill (option 2 of [[2026-07-28-verification-beat-design]]) is added later, this beat should move earlier so the problem precedes its solution.
|
||
- **Stage-safety note added inline** (`_note:`): a modern model may disambiguate "right shelf" correctly from context, so this beat must be pinned to a deterministic response or a low-temp on-rails prompt. Consistent with the Plan's existing "never a naked live call" production rule. Without pinning, the demo can silently succeed and kill the point on stage.
|
||
- Effect on the wiki: this is the first *demo* of [[leave-less-room-for-imagination]] in the deliverable — thesis T12 previously had no dramatization. The script's own accidental ambiguity became the example.
|
||
- Pages changed: [[2026-07-28-webinar-theses]] — added a legend explaining the `T<number>` shorthand (T = thesis; stable handles for cross-referencing; T1–T15 follow v1's order where the thesis survived; letter suffixes mark v2 additions placed beside their nearest relative instead of renumbering). This was an undocumented convention I introduced in v2 and the user was right to flag it.
|
||
- Next unchanged: decide on the checker skill (option 2) and the closing what-stays-yours half from [[2026-07-28-verification-beat-design]]; remaining lint recommendations stand (a `taste` concept page; extend [[theo-konstantin-allie]] to four lenses).
|
||
|
||
## 2026-07-29 — ingest (А что если наВайб-Кодить / "What if we vibe-code it?")
|
||
- Intent: ingest
|
||
- Input: `raw/sources/А что если наВайб-Кодить.md` — viewer's conclusions from a 5:31 Russian YouTube video (author unknown; his company pays "millions a year" for Datadog). 11th source; `raw/sources/` fully ingested again.
|
||
- Pages created: [[2026-07-29-what-if-we-vibe-code-it]] (source), [[maintenance-is-the-real-cost]] (concept — 25th).
|
||
- Core of the source: **writing code was never the bottleneck — maintenance is.** Developers never skipped building their own Jira/Datadog for lack of ability; they skipped it because they didn't want to *run* the result. An internal service is a second IT business (bad for the company and the developer both); "I can write it in a week" ≠ "worth writing"; the vendor sells operational offload, not code. Evidence: the pendulum case — a company builds its own Jira clone (March 2026) and returns to a bought tracker, Linear, by July. Prescription: a build-vs-buy checklist (dependency size / ongoing support / operational load / second-business willingness). The author also retracts his own earlier "many services will die because of AI" claim.
|
||
- Why it matters to this vault: it is the **first dedicated counterweight to the build-everything-yourself thread** ([[explosion-of-internal-software]], [[emacsification-of-software]]) — and both of those pages had already flagged "maintenance is assumed away" as their own weakest point, so the objection was latent and is now sourced with the corpus's only *observed outcome* of the pattern (a reversal). Reconciliation recorded on both sides: Thorsten's club app *passes* the source's own checklist (tiny, personal, no SLA), his "teams will remix Riverside" prediction is what the checklist rejects — the disagreement is a threshold, not a winner. Separately, the source *agrees* with the vault's spine from a new angle: "writing was never the bottleneck" is the harness-over-model premise; the corpus now holds three named bottlenecks that don't compete — context (Thorsten, authoring time), verification (Theo, ship time), maintenance (this source, lifetime).
|
||
- Pages updated: [[explosion-of-internal-software]] (contradiction upgraded from self-criticism to sourced, threshold reconciliation, next-question sharpened), [[emacsification-of-software]] (maintenance objection sourced; `~/bin` fork passes, team remix doesn't), [[code-as-throwaway]] (new "lifetime boundary" bullet — throwaway is safe *because* unmaintained; first user converts code into a service), [[thorsten-ball]] (contradiction added, tentative both sides — one anecdote vs one prediction), [[overview]] (11 sources; frontier bullet counterweight; new divergence entry; concept count), [[2026-07-28-webinar-theses]] (T5 scoping note: the thesis survives — its examples are checklist-safe — and gains a one-sentence inoculation against the sharpest technical-audience pushback), `index.md`.
|
||
- Uncertainty flagged: the pendulum case is second-hand tweets with fuzzy company identification (tentative); the checklist is prescriptive, not observed; the author's own company is currently building a Datadog replacement — if it ships and survives, he becomes his own counterexample. Open question with no evidence either way in the corpus: does the maintenance objection survive *agents* doing the maintenance ([[agentic-loops]], [[async-by-default]])?
|
||
- Effect on the last lint's watch item: [[emacsification-of-software]] and [[explosion-of-internal-software]] were merge candidates "if neither gains second-source support by the next lint" — both now have second-source engagement (as a bounding counterpoint), which argues for keeping them separate with [[maintenance-is-the-real-cost]] as the shared boundary page.
|
||
- Next: the standing recommendations are unchanged (a `taste` concept page; extend [[theo-konstantin-allie]] to four lenses; the skills falsification test). New candidate question for the HR audience: which of their candidate tools (candidate knowledge base, transcribe→summarize) fall on the safe side of the build-vs-buy checklist — directly webinar-relevant if Q&A raises "should we build or buy?"
|
||
|
||
## 2026-07-30 — query (Stanford "widening gap" chart traced to source)
|
||
- Intent: query
|
||
- Input: user saved `raw/assets/G6g3O60bkAE05ZW.png` (X/Twitter screenshot of a Stanford slide, "Teams that master AI are accelerating their productivity gains, widening the gap with laggards") and asked to find the original research.
|
||
- Pages created: [[2026-07-30-stanford-widening-gap-source]] (query — the slide is from Stanford's Software Engineering Productivity Research group (SWEPR), Yegor Denisov-Blanch; 600+ companies / 120k+ engineers since 2022; the chart is a 46-vs-46-team difference-in-differences analysis showing the AI-adopter productivity gap growing 4.8% → 19% (4×) from April 2023 to July 2025; slide matches his Sept 2025 AI Conference deck "Will AI Replace Software Engineers?"; primary links recorded on the page).
|
||
- Pages updated: [[2026-07-14-gap-between-ai-users-irreversible]] (external-corroboration pointer under Connections — Allie's title claim gains its first measured, non-practitioner support), `index.md` (Queries).
|
||
- Notes: the DiD result itself is talk/deck-published, not peer-reviewed — marked tentative; only the measurement methodology has a peer-reviewed paper (arXiv 2409.15152). Nuance recorded: same study finds ~15–20% average gains with AI *decreasing* net productivity in complex legacy codebases (rework, +91% PR review time) — honest-caveat material aligning with [[maintenance-is-the-real-cost]] and [[make-more-cheap-code]]. No concept pages changed (citation policy: the talk is not yet an ingested source).
|
||
- Next: decide whether to ingest the talk/deck as a proper `raw/sources/` doc (would let concept pages and the webinar theses cite it as `wiki/sources/*` evidence); consider adding the "Stanford measured it: 4× in two years" line to the webinar's stakes beat; watch for a peer-reviewed version of the DiD analysis.
|
||
|
||
## 2026-07-30 — ingest (Stanford SWEPR — AI and the widening productivity gap)
|
||
- Intent: ingest
|
||
- Input: user — "Create the source. I'm giving you permission to write a new file in the sources folder." **Authorized raw write** (exception to rule 1, new file only): created `raw/sources/Stanford SWEPR - AI and the widening productivity gap.md`, a research *dossier* compiled from the saved slide screenshot (`raw/assets/G6g3O60bkAE05ZW.png`) plus public coverage — honestly marked as not-a-transcript, with per-claim provenance (slide-read vs secondary coverage). 12th source; the corpus's first quantitative outside study.
|
||
- Pages created: [[2026-07-30-stanford-swepr-widening-gap]] (source), [[swepr]] (entity — Stanford Software Engineering Productivity Research group, Yegor Denisov-Blanch; 15th entity).
|
||
- Pages updated: [[levels-of-ai-usage]] (evidence: measured team-level twin of the mastery gap, with the non-engineer-audience caveat), [[context-as-scarce-resource]] (evidence: gains collapse toward 10M LOC via context-window limits — first outside quantitative support for context-as-constraint), [[make-more-cheap-code]] (evidence: +91% PR review time / 2.6× rework = the cost moving downstream, measured; brownfield-negative caveat), [[2026-07-14-gap-between-ai-users-irreversible]] (corroboration pointer upgraded to cite the source page), [[2026-07-28-webinar-theses]] (T7 note: thesis upgraded from prediction to measurement, with a citable stage line), [[2026-07-30-stanford-widening-gap-source]] (ingest follow-up marked done), [[overview]] (11→12 sources; stakes-claim measurement added to the agree section; navigation), `index.md`.
|
||
- Notes: the source *corroborates* rather than contradicts — no new contradictions logged, but three caveats recorded on the source page and carried into every citing page: the DiD analysis is talk-published, not peer-reviewed (tentative); the rework/review numbers are secondary coverage; the study measures engineering teams, not the webinar's non-engineer audience. Deliberately did **not** claim the study validates the *skills* rung specifically — Stanford asserts "quality of usage" as the differentiator without decomposing it, so it cannot arbitrate contradiction #1 (skills vs no-skills). The webinar theses' T7 gains the corpus's best number: gap 4.8% → 19% (4×), April 2023 → July 2025.
|
||
- Next: watch for a peer-reviewed version of the DiD analysis (status upgrade); candidate webinar edit — the one-line "Stanford measured it: the gap quadrupled in two years" in the stakes beat (user's call, script is an authored deliverable); standing recommendations unchanged (a `taste` concept page; extend [[theo-konstantin-allie]] to four lenses; the skills falsification test).
|
||
|
||
## 2026-07-30 — maintenance (authorized raw edit: Stanford beat in the script's closing arc)
|
||
- Intent: maintenance
|
||
- Input: user — "good idea to add it to the webinar script. Do it."
|
||
- **Authorized raw edit** (exception to rule 1, on direct instruction; same pattern as the 2026-07-14/07-28 script edits): `raw/notes/Webinar script.md` — inserted a Stanford beat into the closing arc, between "You don't buy it. You build it — one small tool at a time." and "We started this journey…". The beat: Stanford tracked 46 AI teams vs 46 matched non-AI teams for 2+ years → the teams that *learned* it pulled away from the ones that just *had* it → spread under 5% (spring 2023) → 19% (summer 2025) → "the gap quadrupled in two years" → callback to the script's own reveal: "everyone had the same models the whole time. The difference was never the model. It was who built something around it."
|
||
- Placement rationale: the closing arc is where the script's "same model, different harness" argument lands, and the Stanford curve is that exact argument as data — it also answers "why start now" right before the final chat-box→OS callback. Phrasing kept factually careful: the widening spread is *among AI-using teams* (masters vs laggards), so the beat says "pulled away from the ones that just had it," not "AI users vs non-users."
|
||
- Stage-safety/honesty note added inline (`_note:`): source pointer to [[2026-07-30-stanford-swepr-widening-gap]] plus the three Q&A caveats (talk-published not peer-reviewed; software teams not office workers; "quality of usage" asserted but not decomposed).
|
||
- Pages changed: [[2026-07-28-webinar-theses]] (T7 note updated — dramatized in the script as of today, promoted from Q&A material to an on-stage beat). No other wiki pages changed; `index.md` unchanged (no catalog change).
|
||
- Note on scope: the Stanford research itself was already fully ingested earlier today ([[2026-07-30-stanford-swepr-widening-gap]] + [[swepr]] — see the previous ingest entry); this operation only carries the number into the deliverable.
|
||
- Next: unchanged from the ingest entry (peer-review watch; `taste` concept page; four-lens comparison; skills falsification test).
|
||
|
||
## 2026-07-30 — ingest (Грабли во внедрении ИИ в SDLC / Nikolai Sheiko)
|
||
- Intent: ingest
|
||
- Input: `raw/sources/Грабли во внедрении ИИ в SDLC.md` — viewer's conclusions from Nikolai Sheiko's 45:59 Russian YouTube talk ("why the AI is there but the results aren't"). 13th source; `raw/sources/` fully ingested again.
|
||
- Pages created: [[2026-07-30-rakes-in-ai-sdlc-adoption]] (source), [[nikolai-sheiko]] (entity — 16th), [[review-is-the-new-bottleneck]] and [[developer-as-agent-manager]] (concepts — 26th and 27th).
|
||
- Core of the source: models are already good enough — **people, companies and metrics throttle the gains by an order of magnitude**. The SDLC collapsed into days/hours but not around the two human "red squares" (reviewer, planner); the fix is review *with* the agent plus the one metric that resists gaming — **completed tasks without rework** (never LoC/commits/PRs; his outsourcer case: more PRs, +1% net). Error #0: the AI-developer is an IO-bound **manager of an agent-employee**, not a CPU-bound coder ("sit watching Claude Code work = bad employee") — and not everyone can or should switch. "Companies no longer need custom AI development — install Claude Code/Codex, configure, attach connectors, mind security." **Agentic Evolution**: walk the agent through hard tasks → "remember this and write the manual for the next one" → verify via a **context-free subagent** solving the task from the skill alone. Compaction curse on big codebases → best practices exist *for the agent* (locality, interfaces, AST search over grep); embeddings/RAG over code rejected flatly.
|
||
- Why it matters to this vault: (1) **second practitioner vote for the skills layer**, narrowing the evidential asymmetry Thorsten's dissent enjoyed on [[skills-as-memory]] — and his verification protocol is the corpus's first described run of anything like the proposed skills falsification test (with-skill half only; no without-skill control, so the test question stands). (2) **Independent citation of the Stanford chart** ([[swepr]]) as his stakes slide, plus an anecdotal mirror of its +91%-review-time finding — recorded on [[2026-07-30-stanford-swepr-widening-gap]]. (3) Names the org-level bottleneck the corpus had only as an open question on [[async-by-default]] (parallel diffs pile up on a human) — now a page: [[review-is-the-new-bottleneck]]. (4) His anti-RAG-for-code stance supports the audience-driven reading of the skills-vs-RAG contradiction on [[context-as-scarce-resource]] (anti-RAG votes are about code/procedures; the pro-RAG vote is about business data).
|
||
- Pages updated: [[solve-first-then-skillify]] (Agentic Evolution + 5-step verification protocol; next-question partially answered), [[skills-as-memory]] (evidence + asymmetry softened + falsification-test note), [[context-as-scarce-resource]] (compaction curse; codebase-stores-context converging with Thorsten from the opposite direction; AST search; RAG contradiction note), [[async-by-default]] (IO-bound-manager evidence; related links to both new concepts), [[enterprise-ai-reality]] (no-custom-AI-dev quote as the managed-harness market seconded; external-configurator vs teacher/curator anti-pattern; Cursor metered billing → team economizes, the metered-vs-subscription split observed organizationally; tokens-dearer-then-cheaper matching Eugene's prediction), [[make-more-cheap-code]] (related link to the review-bottleneck page), [[2026-07-30-stanford-swepr-widening-gap]] (independent-citation pointer), [[overview]] (12→13 sources; new "adoption side" bullet; agree/diverge updates; skills-dissent paragraph rebalanced), `index.md`.
|
||
- New contradiction logged (on [[developer-as-agent-manager]]): Sheiko's "don't force everyone" vs the irreversible/compounding gap (Allie, Stanford) — opting out is legitimate *and* costly; no source reconciles the two. Also tentative: all client cases are anonymous self-reported anecdotes; "embeddings over code don't work" has no mechanism given; the users-vs-Agentic-Operations split is prediction, not observation.
|
||
- Uncertainty about the speaker himself: affiliation unknown; "you don't need custom AI development" is also a consultant's pitch — flagged on [[nikolai-sheiko]].
|
||
- Webinar relevance noted but not applied (script untouched): the Intelligence-vs-Judgment framing is close kin to the planned "what stays yours" closing beat, and the talk independently strengthens the case for the standing `taste` concept-page recommendation (Judgment = taste-built-over-years or domain expertise — a second source alongside Thorsten's).
|
||
- Next: standing recommendations unchanged (a `taste` concept page — now with two sources backing it; extend [[theo-konstantin-allie]] to four lenses; the skills falsification test — half-run by this source, control still missing). New candidate question: what does review-with-the-agent look like concretely (no transcript of it done well exists in the corpus).
|