ASSAY Try one repo

Conference overviews · Grok Bot Galaxy 2026 17 event recordings + 1 adjacent talk

assay · overviews · 2026-09-15–17 · livestream

Grok Bot Galaxy 2026

We reviewed 17 event recordings plus one adjacent standalone talk around Grok Bot Galaxy, the 15–17 September 2026 build-and-workshop livestream. The recordings are third-party re-uploads, and two event segments were unavailable in the reviewed set. Findings update the concept register, including one new concept: demonstration-to-skill capture.

Reviewed 2026-10-02 · Grok Bot 101 recording · adjacent Lauren Tan talk

§1The event in one paragraph

The live build and role-specific workshops show named bot fleets coordinating coding, product, sales and support work. Their value is partly in the inconsistency: a support knowledge-base workflow retains a human code owner, while a disposable game prototype deliberately relaxes review and experiences defects and an outage. The adjacent Lauren Tan talk argues for building trust through verifiable behaviour. These accounts reveal different risk choices; they do not establish one uniform production policy.

§2The 6 learnings that matter

  1. A demonstration can author a skill

    A human records a task on the bot’s computer and the product turns it into a reusable procedure, sometimes with voice narration. Four accounts of one product still count as one independent source. See demonstration-to-skill capture.

  2. Specialists need a coordinator and visible handoffs

    The workshops repeatedly separate bot roles and context, then place a coordinating bot between the specialists and the human. The demonstrations establish a pattern within one vendor, not several independent confirmations. See multi-agent organisation design.

  3. Review boundaries differ by workload

    Support changes pass through a PR, automated review, branch evals and human approval. The prototype game uses automatic merging and additional testing, but defects still escape. Neither experience alone establishes the right boundary for another system. See layered merge gates and graduated authority.

  4. Verification needs an executable interface

    A reusable CLI and feature map let agents exercise the application instead of inferring correctness from code. Play-testing and a second-stage triage validator broaden the checks, but their defect-escape rate is unmeasured. See independent verification.

  5. The compute boundary needs verification

    One workshop describes one inaccessible VM per bot; another describes instances sharing a VM and files. Preserve that conflict until architecture evidence resolves it. See per-agent isolation and filesystem memory.

  6. Attention and cost remain finite

    The bot fleet identifies serial human merging as a constraint. Other sessions reduce routine notifications and discuss batching support work or replacing polling with events. Ticket-cost figures are presenter reports, not audited service benchmarks. See human attention and cost per outcome.

§3Where Assay stands

Assay’s separate roles, bounded work items and independent verification address several problems visible in the demos. The self-merging prototype is a counterexample to universal human review, not a reason to silently widen an adopting project’s authority. Assay does not describe a demonstration-recording skill authoring path.

§4What this event added to the register

The scan contributes to 20 existing concepts and adds demonstration-to-skill capture, bringing the register to 65 concepts across 14 themes. Evidence includes corroboration, unresolved architecture descriptions and explicit counterexamples. Multiple talks from the same vendor count once toward a concept’s independent-source total.

§5Method and coverage

Eighteen per-segment recordings were reviewed from automatic English subtitle transcripts: 17 event recordings and Lauren Tan’s adjacent standalone talk. They come from the third-party Grok Bot Galaxy channel, not the organiser’s own channel. Day 3 Part 02 and Part 05 were absent; no substitute was reviewed. Full-day streams and creator recaps were excluded from this pass. The adjacent talk was first posted on 2026-09-21 and re-uploaded on 2026-09-22; its delivery date is unverified and it is not dated to the event. Individual workshop dates within the three-day event are not established. Demos, reported results and independently verified results are kept distinct.

Every conference sighting in the register links the recording used for the review.