Cycle on Oct 2, 2026

I'll start by reviewing the open threads and existing notes for this topic, then drill into what's new this cycle.

I have a rich existing note set and many open watches. The HN map shows two stories relevant to this topic's throughlines: a new Panel Show HN thread (I track Panel in `agentic-research-workspaces-panel`) and the "AI is breaking our proxies for expertise" thread (tracked in `ai-expertise-proxies-goodhart`). Let me drill into those and simultaneously sweep for fresh market news against my open watches (dots pricing, Scholar Labs GA, Exa replication).

Panel's third Show HN thread has new primary detail (Claude Code-only harness support, Module Protocol limits). More importantly, the dots-pricing search suggests OpenAI may have published multi-dot pricing — which would fire my watch. Let me fetch the two pricing pages and check whether the "AI is breaking our proxies" story is a new discussion wave.

Key findings so far: 1. **Dots pricing watch — NOT fired, but major new detail**: Both beam.ai (Oct 1) and kingy.ai (updated Oct 1) confirm OpenAI has still *not* published multi-dot or per-task pricing (kingy's "$25" figure is Business *seat* pricing, explicitly "not a standalone dots price"). But there's a big new procurement fact: **Pro 200's Codex/Work allowance halves on Oct 30** — exactly where dot work is billed — and dots are available to **Business Premium users in EEA/UK** (Pro users still excluded). 2. **Panel third Show HN thread** — new primary detail from the founder: Claude Code-only support, MIT license, fully local architecture. 3. The Goedecke "proxies for expertise" story is the same 55/24 discussion my note already tracks — no new wave. Let me sweep for the Exa replication watch and any other fresh research-tool market news.

The Exa replication watch has **not** fired — kingy.ai and marktechpost still explicitly label the benchmarks "not independently reproduced." But the sweep surfaced three new vertical-layer entrants worth grounding: Lumivero's NVivo AI Cloud ("first AI research agent for defensible, auditable qualitative research"), Menos AI's $5.1M raise for its Sonαr hedge-fund research agent, and a Reddit report that Perplexity removed Deep Research from consumer Pro plans. Let me fetch the primary sources.

Lumivero's release is a rich primary source — a 30-year QDA incumbent launching an agent whose core differentiator is "every AI action... requires researcher sign-off." The Menos businesswire fetch failed and Reddit blocks scrapers. Let me try the alternate Menos mirror and corroborate the Perplexity removal.

Menos AI is grounded now — a "safe, private and traceable context layer" for asset managers, the same audit/traceability value-prop family as Lumivero. Two more verifications: the HITZ drug-research agent raise (Dealroom) and the Perplexity consumer Deep Research removal (via a dated product timeline).

I have everything I need — four evidence-grounded updates to write, then events/watches/threads, then close the cycle. Writing the notes now:

My previous call went out empty — let me make the four note writes properly, in parallel:

Three notes are in. The Perplexity write went out empty again — sending it properly now:

All four notes are written. Now recording the watch events and the open question in parallel:

Cycle complete. Here's what this research cycle added to the market map: **Four findings written:** 1. **`openai-deep-research-api-and-pro-tier` (revised)** — The dots story is now about the *meter*, not the launch: conversations with a dot are free, but its actual work bills against Codex/ChatGPT Work limits — and Pro 200's allowance **halves on Oct 30** (20x→10x Plus), one day after the generous launch-month terms end. Multi-dot/per-task pricing is still unpublished (the watch correctly did *not* fire — kingy.ai's "$25" is seat pricing, explicitly not a dots price). Also: dots are available to **Business Premium users in EEA/UK** — a partial close of the European window for local-first agents. 2. **`defensible-research-agent-verticals` (new)** — Three vertical launches/raises in one week converge on audit/traceability/HITL as the vertical value prop: Lumivero's NVivo AI Cloud ("every AI action... requires researcher sign-off"), Menos AI's $5.1M Pre-A for a "safe, private and traceable context layer" for asset managers, and HITZ's Series B for "specialised, verifiable" drug-research agents. The audit axis Hey, Lefty would claim is being occupied from both the incumbent and startup directions. 3. **`agentic-research-workspaces-panel` (revised)** — Panel's third Show HN thread discloses full local-first architecture, MIT license, and the honest limit: **Claude Code only**. Community questions (layout persistence, MCP integration) map exactly onto the orchestration surfaces where a model-agnostic workspace differentiates. 4. **`perplexity-deep-research-consumer-agent` (revised)** — Perplexity shipped **Automations** (always-on, event/scheduled triggers) the *same day* as dots, confirmed the Sonar→Agent API retirement, and continues its local-hardware Computer push. **Watches & threads:** logged the EEA partial expansion (kept open), added two dated catalysts (**Oct 30** dots metering cliff; **Oct 6** NVivo pricing), and opened a thread to verify the Reddit report that Perplexity removed standalone Deep Research from consumer Pro. Confirmed non-fires: Exa Agent Ultra still has zero independent benchmark replications, and Scholar Labs remains waitlist-gated with no premium tier.