Its own framing, from the homepage:

There's no other event like this -- a conference for engineers by engineers,
where we'll share novel approaches for solving complex problems efficiently
and at speed. Vendor and tool agnostic, this conference will be for a highly
technical audience only. Your boss's boss is not invited.

The CFP adds the constraint that makes the vendor-neutrality claim testable: "It's vendor-neutral, and we don't allow shallow overviews or product pitches." Talks are 15–20 minutes.

A conference about tail latency has put coding with agents in both day-one keynote slots and the day-two opener. That is the finding, and it is about where the subject has landed rather than about any single talk.

• Effective Coding and Working with Agents — Chip Huyen, author of AI Engineering. On making "the most out of working with multiple agents".
• Hardwood: Building a Parquet Parser From Scratch (With a Little Help From AI) — Gunnar Morling, Confluent. The abstract names the tool: "practical learnings from using AI (specifically, Claude Code) as a coding companion".
• How to Improve Your Cache Algorithm Using AI — Dor Laor, ScyllaDB's CEO, opening day two. "Everybody knows that AI can code. It can also convert your code to Rust. But can AI write a database? … my journey of 'vibe coding' the ScyllaDB cache to implement an alternative LRU algorithm and improve our p99 latency."

Roughly eighteen of sixty-two talks touch AI. Two are close enough to the work here to be worth the timeslot.

61 talks from the homepage featured-speaker list, read 2026-10-08, co-presenters merged. Day and slot per talk are not recorded here: the list carries none, and the full agenda grid has not been re-read since 2026-09-26. The agenda counted 62; the one-talk gap is unreconciled.

The Related here column links a note on this site. A plain link means the note covers the talk's named subject. "guess:" means the overlap is inferred from the title alone. A blank means nothing here is close.

The talks with the most overlap with work here, beyond the two agent-in-production talks above. Where an abstract has not been read, the entry says so and stops at the title.

Each talk was matched against the published notes under site/research/, site/events/ and site/tools/, searched by title, description and keywords and then read. Abstracts are still unread for most talks, so a match is against the talk's stated subject. Strong means the note works the talk's central subject; partial means it covers a named part of it; tangential means the overlap is a single component or a passing mention.

Thirty-two of 61 talks match: eight strong, thirteen partial, eleven tangential. Networking fits worst, with nothing for any of its six talks. The largest gap is I/O and storage: seven talks, two of them keynotes (Kivity on the I/O request path, Kreuzmayr on object storage), plus Axboe and Costa on io_uring, and the site has no research on io_uring, storage engines or object-storage clients, only a RustConf listing. Carl Lerche's keynote and the runtimes subsection come next: nothing here covers async runtimes or schedulers beyond a queueing simulator.

For Every View Is Late, Tracegrams is prior art: a summary that keeps tail behaviour across hops is the merged-distribution answer the proposal asks for. From Spans to Answers is the counterpoint, keeping more and aggregating at query time. P(od)99 and Closed-Loop Performance Engineering bear on the load-model half; whether either distinguishes open- from closed-loop measurement is worth asking in Q&A. The TSDB and precise-time talks touch the scrape-age and interval-clock examples.

p99conf.io/on-demand carries 264 session pages spanning 2021 through 2025, filterable by year, type and topic, each with abstract, speaker bio and slides. Ungated — an attendee quote the site chooses to display reads "Videos available on-demand afterwards with no gating or games."

For a conference whose talks are recorded in advance anyway, the archive rather than the live event is the artifact. ScyllaDB also runs a separate gated on-demand landing page for 2025, which is worth knowing before handing over an email.

Recorded because a reader planning around this page would hit them.

• The UTC conversion is wrong. The site says "8:00am – 1:00pm Pacific Time / 16:00 – 20:00 UTC". October Pacific is UTC-7, so 08:00 PT is 15:00 UTC, not 16:00. Sessionize's own timezone field says UTC-07. Unchanged on 2026-10-08: the page still pairs 08:00–13:00 Pacific with 16:00–20:00 UTC. Pacific Daylight Time makes it 15:00–20:00 UTC; the end is right and the start is an hour late. Recorded as found.
• The stated hours exclude a whole track. "8:00am – 1:00pm" on both days, while the grid schedules nine day-two sessions from 01:00 to 03:55 Pacific for Europe, Asia and India. Day two actually runs about 01:00 to 12:35.
• "Full agenda will be announced soon" still sits on the homepage while /agenda/ renders a complete two-day, three-stage grid.
• Three talks carry different titles on the homepage and in the agenda, including one where the vector count differs — 1B on the homepage, 100M in the abstract.
• The lower speaker wall is a legacy reel. Andy Pavlo, Michael Stonebraker, Gil Tene, Liz Rice, Charity Majors and others appear with bios and no 2026 talk. They are past speakers used as social proof and read as this year's lineup.
• One talk exists only on the homepage — Catherine Jue of KERNEL on "building crazy fast, open source infrastructure for AI agents", with sub-30ms browser boot times — and appears nowhere in the agenda feed. It is still on the homepage list on 2026-10-08; the feed was not re-checked.
• 61 featured, 62 scheduled. The homepage list read on 2026-10-08 has 61 talks against the 62 the agenda grid held on 2026-09-26. Which one is missing from which list is not established.
• The keynote lineup does not match this page's framing. The organiser email of 2026-10-20 names eight keynotes, four a day (see Keynotes). Chip Huyen's Effective Coding and Working with Agents is not among them, so "both day-one keynote slots" being about agents, as the title and Why this one matters here claim, is not supported by the email. Of the four day-one keynotes, only Gunnar Morling's names AI. Whether Huyen moved to a regular slot or the email omits her is not established. The title is left as written.
• Two affiliations differ. The talk list carries Carl Lerche as AWS; the email calls him the creator of Tokio. This page calls Dor Laor ScyllaDB's CEO; the email calls him co-founder. Neither excludes the other.
• "70+ speakers" against 61 talks. Not a conflict: the 61 talks in the list carry 70 named speakers once co-presenters are counted. The 14 unscheduled talks in the Event table date from the 2026-09-26 grid read and have not been rechecked against the email.

Direct: the note covers the talk's named subject, and carries a Seen at link back.

• Agent Sandbox Architectures — Lovable, Unikraft: sandbox lifecycle and boot cost.
• Performance Engineering Tools Reference — Den Odell, whose book it maps; Tanel Poder and the other eBPF talks; Tracegrams and Salesforce on distributed tracing; HERE and Stax on profiling; the Valkey talk against its caching section.
• The Oracle Is the Deliverable — Netflix: canary results, not model confidence, as the oracle an autonomous agent is graded by.
• Context Surfaces — Redis: profiles exposed to an agent over MCP.
• Python Priority Queues with heapq — the SimdQuickHeap talk in the databases table.

One-way, because crowsnest is a versioned spec rather than a research note:

• crowsnest OTLP ingest — Tracegrams, Salesforce. The receiver stores every span; both talks are about not doing that at scale.

Read for the alignment pass on 2026-10-20, with no back-link yet:

• Forgetting and Attribution: cache replacement — Dor Laor's AI-written cache algorithm. The note scores eviction against Belady's MIN; an LRU alternative claimed to improve p99 needs that score.
• Every View Is Late — Tracegrams as prior art, Salesforce as counterpoint, P(od)99 on the load model. A P99 CONF 2027 proposal written against this program.

Guesses, inferred from a title or a one-line abstract, with no back-link:

• Agentic Software Engineering — Chip Huyen.
• The Oracle Is the Deliverable — Perfloop, Disney+/Hulu's AI-driven optimization, Dor Laor's AI-written cache.
• Rebuilding from the spec — Gunnar Morling's Parquet parser built with Claude Code.
• Context Surfaces — The Prompt is the Platform.
• Agent Sandbox Architectures — KERNEL.
• Sandboxing AI Coding Agents with FreeBSD Jails — Lovable; a jail lives per agent, a Lovable cell per prompt.
• Serverless Architecture — Unikraft, on cold starts.
• Agent Telemetry Systems — Tracegrams.
• TLA+ for System Design — American Express, and AWS on optimistic concurrency and precise time.
• pocket-es — Vectorless RAG and the ScyllaDB full-text benchmark; pocket-es is BM25 with no vectors.
• Qwen3.6 and the KV Cache Constraint — Fearless Quantization.

Rejected after reading: bac and P-99 (the name is Ninety-Nine Prolog Problems, not a percentile); caching (HTTP and CDN caching, not eviction algorithms); functools (lru_cache usage only); Go (no execution-tracer content); racket-treelist, faster-simpler-red-black-trees and representing-sets (no hashing or heap content); agent-sandbox-practical-configs (jail and Seatbelt configuration, not boot cost). Nothing here covers io_uring, eBPF beyond the tools reference, unikernel images, Parquet, QUIC or vector indexes.

• p99conf.io — server-rendered; the homepage carries the speaker list and abstracts
• CFP — closed 29 May 2026; carries the track list verbatim
• On-demand archive, 2021–2025
• 2025 recap — "the fifth in this incredible series", titled Latency to LLMs