Cycle on Oct 5, 2026

I'll run this research cycle on the AI research tools market. Starting by reviewing open threads and existing notes, then drilling into the HN stories relevant to the brief (Panel — a research workspace — and the "AI is breaking our proxies for expertise" piece both touch the market directly).

I have a well-developed note set — the most recent revisions are Oct 2–4 (Panel workspace, Perplexity's triple launch, the Goodhart/expertise-proxies thread). The HN map's two on-brief stories (Panel at 22 pts, "AI is breaking our proxies for expertise" at 55 pts) match what those notes already cover. So this cycle's highest-value work is: (1) run the cheap tripwire checks on open watches (Perplexity Enterprise pricing, Scholar Labs GA, Exa Agent Ultra independent replication), and (2) sweep the last week for market developments the notes don't yet have. Running those checks in parallel now.

Good signal in the sweep. Four things stand out: (1) a **Gemini 4 Argon** launch post on blog.google with an "introductory price" — new since my notes; (2) the **Gemini API pricing page now has a "Pricing for agents" section** selling the Deep Research agent as a metered API primitive; (3) Perplexity's Enterprise **credits help-center articles** are live — need to check if actual dollar prices are published (that's the tripwire); (4) an **agentic.ai market map of research agents, updated October 2026** — a third-party map to cross-check. Fetching all four in parallel.

I have all the evidence I need from this cycle's four fetches. Now writing the three finding revisions in one batch — (1) Gemini Deep Research note updated with the Argon launch + "Pricing for agents" metering, (2) Perplexity note updated with the Enterprise credit mechanics, (3) market map updated with agentic.ai's third-party rankings — then registering watches and closing the cycle.

All three notes are saved. Now registering the two forward-looking watches this cycle surfaced: Argon's general availability (the rollout is explicitly phased, no date given) and the scheduled Jan 1, 2027 doubling of Gemini 3.x Flash list prices.

Both watches registered. The cycle is complete — calling finish_cycle.

Cycle complete. Here's what this cycle did: **Checked the open tripwires** (all quiet on three): - **Perplexity Enterprise pricing** — partially advanced: the new credits admin guide publishes the *mechanics* (Enterprise Pro 500 credits/mo, Enterprise Max 15,000, two-pool system, named usage limits replacing seat-tier caps) but still no dollar price per credit → watch stays open. - **Scholar Labs GA/premium** — nothing new; still waitlist-gated per the UNC library guide. - **Exa Agent Ultra independent replication** — sweep found zero neutral evals; only SEO republishers → watch stays open. **Three findings written** (revisions to existing slugs): 1. **[[gemini-deep-research-agent]]** — Google announced **Gemini 4 Argon** (Sept 30): 1M-token output limit (up from 64K), intro pricing $2/$10 per million tokens, benchmarks aimed squarely at research-agent workloads (Vals Finance Agent v2, Harvey's Legal Agent Benchmark), and a phased rollout that hasn't reached the public yet. Separately, the Gemini API pricing page now has a **"Pricing for agents"** section: Deep Research is billed as *pass-through token metering* — including intermediate reasoning tokens — with no per-run premium. Google is pricing deep research as a commodity primitive, compressing the umbrella Exa and Parallel sell under, and its meter is the only fully public one (vs. OpenAI's unpublished dots allowances and Perplexity's unpriced credits). 2. **[[perplexity-deep-research-consumer-agent]]** — the Enterprise credit disclosure above, plus the pattern note: Perplexity and OpenAI both published meters before prices, while Google publishes price directly. 3. **[[market-map-positioning-hey-lefty]]** — agentic.ai's October 2026 third-party **"Agenticness" rankings** (ChatGPT Deep Research 17/36 top score; Consensus 4/36; STORM 2/36): no destination tool scores as fully autonomous, buyers are coached that verifiability is non-negotiable, and the list is destination-layer only — all three points support Hey, Lefty's audit-grade orchestrator positioning. **Two watches registered:** Argon general availability (paid API / AI Ultra), and the scheduled **Jan 1, 2027 doubling of Gemini 3.x Flash list prices** — a 2x COGS increase for every research tool built on Gemini Flash.