AI Native DevCon 2026 New York — practical agent skills grounded in conference talks
92%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Talk: Context != Context: One Window, Different Lifecycles
Speaker: Birgitta Boeckeler
Event: AI Native DevCon New York 2026
Recording: https://www.youtube.com/watch?v=NxypyNu1944
This index paraphrases the supplied corrected SRT. The accompanying transcript retains its original cue numbers, full start/end timestamps, and spoken wording, with subtitle line wrapping joined. Cues 195–205 contain an inserted conference promotion and are omitted. The unchanged timing allows references to the supplied recording.
| Topic | Timestamp | Cues |
|---|---|---|
| Why one context window contains semantically different material | 00:27–02:18 | 8–38 |
| Codified and ambient context: authorship, reuse, rate of change | 02:20–04:51 | 39–76 |
| Complementary review rules and incident history | 04:52–05:51 | 77–91 |
| Task phase and repository context; onboarding example | 05:53–07:20 | 92–111 |
| Codified loop attributed to Patrick Debois | 07:21–08:12 | 112–123 |
| Clarity, actionability, fidelity, efficiency, security | 08:33–09:54 | 129–148 |
| MCP delivery: who handles quality, permissions, recall and precision? | 09:56–11:19 | 149–170 |
| Context engine and task-time reasoning | 11:20–12:37 | 171–189 |
| Ambient loop: ingest and assemble replace generate and distribute | 12:38–12:59 | 190–194 |
| Promotional interlude omitted | 13:00–13:37 | 195–205 |
| Ingest: synchronization, freshness, permission models, identity | 13:41–16:00 | 206–240 |
| Evaluate: scenarios, gold documents/answers, conflict and absence | 16:02–17:29 | 241–260 |
| Assemble: caller needs, depth, latency and cost | 17:33–18:57 | 261–282 |
| Conflict example and code as a source of reality | 18:59–20:41 | 283–306 |
| Observe: disputed sources, engine bias, downvoted-answer analysis | 20:43–22:45 | 307–336 |
| Diverse implementations; memory, generated wikis, selective indexing | 22:48–24:53 | 337–365 |
| Put stable instructions and changing facts in suitable places | 24:55–25:39 | 366–376 |
| Maintain input knowledge, access and recurring conflicts | 25:41–28:31 | 377–422 |
| Agent-authored context and possible feedback problems | 28:33–30:03 | 423–443 |
Read surrounding cues before using a quote as evidence.
The transcript includes speech artifacts, fragments, and references to slides that are not bundled. The author names and article title around 08:44 are garbled; do not invent the exact citation. The identity example around 15:51–16:00 is also garbled; use the clear explanation in cues 230–237 instead. The suspected-conflict sample and percentages are approximate, and the transcript provides neither raw observations nor independent accuracy measurements. Discussion of Unblocked describes the presentation's context; current capabilities need separate verification. There is no reliable indication that every passage mentioning a product is a general technical prescription.
.tessl-plugin
skills
talk-boeckeler-context-lifecycles
talk-chanezon-gordon-agent-experience-evaluation
talk-meppiel-versioned-agent-contracts