Batch 23: AI-native engineering & coding agents (19 of 20 talks) - #3
Merged
Conversation
…313 talks) 17 of 20 talks from the coding-agents track + two engineering keynotes (published 2026-08-20 → 08-28). Held pending captions: Liguori/AWS and Bond & Ketkar/Uber uReview (YouTube 429 on every attempt), Debois/Tessl (no captions published) — registered as pending in transcripts/README.md. - transcripts + SPIKE extractions for 17 talks; registered in both READMEs - registry: batch-23 reconciliation — evidence table (+21 verification-gap, +19 ai-native-org, +16/−2 harness-over-model, +9 model-not-bottleneck, +8 agent-economy, +7 context-graphs, +7 value-of-judgement; first counter on pat-accelerated-research), a REFRAME recommending a time-dependence qualifier for pat-harness-over-model (Rogge, Bhatawdekar, Krieger, Box), the "review becomes governance" thread, context-as-bottleneck cluster (Uber 40M-entry graph, Qodo, Unblocked), held ledgers; 9 companies / 19 experts / 87 elements; no patterns coined - convert_1719.py + chunk_talks.py: batch 23 stems - graph: seed-full overwrite (18,467 entities; head v29 → v32 incl. FTS rebuild); 313 talks; chunks for this batch follow in a second load (Gemini embedding quota exhausted mid-run) - RUNBOOK/CLAUDE.md: as-of batch 23; deterministic batches 16–23 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HZf9x7aMcaWvWWV332ysY7
…hunks loaded (315 talks) The two talks held for YouTube caption rate-limiting landed on the delayed retry: Liguori (AWS) "Building a Frontier Development Team" and Bond & Ketkar (Uber) "Building uReview". Extracted, added to BATCHES[23] (now 19 stems), reconverted and validated (frag-23: 9 companies, 22 experts, 99 elements, 89 signals, 26 insights, 19 know-hows, 1,225 edges; merge_validate still only the two intentional dangling targets). Graph: second entity overwrite (v35, 18,694 entities) and the batch-23 chunks — 284 for the first 17 talks (v33/v34) plus 30 for these two (v36/v37), embedded in throttled parts after Gemini 429s — with a full-text index rebuild after each merge. Served counts verified: 4,485 entity nodes + 4,207 chunks over 315 talks; top-patterns verification-gap 332 / harness 244 / model-not-bottleneck 178. Docs: transcripts/README and extraction/README move both talks from pending to the batch table (Debois/Tessl stays pending, no captions); registry batch-23 intro, pattern tally (+6 verification-gap, +6 ai-native-org, +2 harness, +2 model-not-bottleneck, +2 judgement, +1 supply-chain), experts/elements lists and an addendum; RUNBOOK's stale "as of batch 21" count block replaced with the batch-23 numbers (head v37). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HZf9x7aMcaWvWWV332ysY7
aaltshuler
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What's in this PR
Batch 23 — the coding-agents / AI-native engineering track from the Sep-3 channel check (18 talks) plus two engineering-practice keynotes (Krieger/Anthropic, Jarmak/Sourcegraph). 19 of 20 processed across two commits:
e675fe0— the first 17 talks (entities loaded, FTS rebuilt).55c6975— the two talks YouTube rate-limited for hours landed on the delayed retry and are now in: Liguori/AWS From AI-Assisted to AI-Native: Building a Frontier Development Team and Bond & Ketkar/Uber Building uReview. Plus all batch-23 transcript chunks (284 + 30, embedded in throttled parts after Gemini 429s), and the RUNBOOK count block refreshed.Graph state (live, verified against the served graph)
Hybrid search returns the new chunks (uReview cavitation, Liguori's 1–2% hands-off coding) and the new signals carry their pattern/element edges.
What the batch says (registry § Batch-23)
pat-verification-gapbatch yet (+27): Lean proofs with humans owning the spec (AWS), review relocation (Qodo; Anthropic shipping intent artifacts — confirming b21's Aviator thesis), evals as an org loop (DoorDash, Braintrust), inner-loop validation (Uber SDLC), and uReview's operator statement — first review latency 3 h → 9 h, agents need a higher accuracy bar than humans (cavitation), expand the outer loop rather than kill it.pat-ai-native-org(+25) gets its best-controlled evidence: Amazon's 50-team, year-long pilot on existing systems — same tools, median 4.5× only for teams that changed how they worked, <3× for the rest; the productivity dip first; flowmaxxing and review load on early-career engineers as the dysfunction side; decision speed as the new bottleneck. Plus Uber 70% of PRs by agents, Anthropic ~60% via Claude Tag.pat-harness-over-model: Hugging Face, Braintrust, Anthropic, Box (b22) and now Liguori (steering files thin as models improve) date heavy orchestration to a model deficit that closed in mid-2025; +18 support, 2 counter-edges kept. Suggested brief qualifier: the harness thins as capability rises; what remains is the deterministic boundary.pat-context-graphs: Uber's 40M-entry context graph with measured token/turn/latency gains, Qodo's software graph, Unblocked's context engine.pat-accelerated-research(Together AI's ParallelKernelBench: 28/87 zero-shot, ~31% plateau) beside two supports (zlib→Lean with 32k lines of proof; models trained overnight by a coding agent).Pipeline notes
Converted deterministically (
convert_1719.pybatch 23, 19 stems): 9 companies, 22 experts, 99 elements, 89 signals, 26 insights, 19 know-hows, 1,225 edges; validation clean (only the two intentional dangling targets). Loads followed the 0.10 procedure (entity overwrite → chunk merge → FTS rebuild → restart) twice. Gemini embedding 429s were handled by splitting the chunk file into 48-line parts with exponential backoff.🤖 Generated with Claude Code
https://claude.ai/code/session_01HZf9x7aMcaWvWWV332ysY7