Autonomous Research Tools: Market Map Digest
TL;DR
The headline this cycle is OpenAI's DevDay launch of dots — always-on agents powered by GPT-6 Astra that absorb deep research as a mode of a persistent host agent, bundled into $200–$500 subscriptions and metered by agent capacity rather than per-report. For the market map, the standalone "deep research report" destination layer just got compressed into the host, and the interesting question shifts to what the host doesn't do — model-agnostic orchestration, audit-grade evidence, and local data control, the last of which is now a hard geographic segmentation because dots are excluded from Europe.
The destination layer got absorbed: deep research is now a mode of an always-on host agent
The product boundary between "deep research tool" and "general assistant" is gone. At DevDay in San Francisco (Sept 29, 2026), OpenAI launched dots — "remarkably capable, always-on agents built to handle everything," each with "their own cloud computer, learn[ing] from feedback over time, and ... work[ing] towards your goals 24/7," connected to "over 4,000 apps" (OpenAI, "Introducing dots"). The New Stack frames the delta as persistence: "Unlike most agents that wait for the next prompt, a Dot keeps working when you step away" (thenewstack.io).
The announcement's own flagship use case is research — a scientist's dot "reruns analyses, investigates unexpected results, and updates the figures for your paper" as new data arrives. Notably, OpenAI built in the research-vs-action firewall that independent orchestration layers have been selling as an evidence standard: background "proactive research" runs through connected apps "restricted to be read-only" (openai-deep-research-api-and-pro-tier). When the host bakes your differentiator into its guardrails, you have to move up the stack — which is exactly what the positioning notes have been arguing.
What to watch: Whether dots' 4,000-app plugin ecosystem becomes OpenAI's licensed-access answer to the gated-web walls documented in the browsing legal precedent thread — that would change the data-access economics for every backend on the map.
Pricing is converging on metered agent work — and OpenAI's top tier jumped to $500
The whole market is standardizing on metered agent capacity, whether sold as subscription allowance or API: Exa Agent Ultra meters per-run swarm budget, Parallel Web Systems meters per-request CPM, and dots are headed the same way — "In the future, you'll be able to add more dots, and scale the output of each dot by either increasing its speed or the total amount of work it can take on per month" (OpenAI). Today, the first dot is included in ChatGPT Pro ($200/mo) and Business Premium ($100/seat), with dot conversations not counting against usage limits (openai-deep-research-api-and-pro-tier).
The second story is a de facto price hike: OpenAI launched a $500/month Pro plan while cutting the $200 plan's allowance in half effective Oct 30 (ChatGPT Work/Codex drops from 20× to 10× Plus), with existing subscribers getting usage credits OpenAI values at $2,500, expiring Dec 31 (The New Stack). Developer reaction was blunt — "Ok guys, back to claude. RIP openAI 🪦" — and The New Stack notes the 20× top-tier allowance had been the standard across labs, with Google AI Ultra and Anthropic's top plan both at $200/mo. The $500 plan's exact allowance remains unpublished.
What to watch: OpenAI publishing concrete list prices for additional dots or per-dot speed/work scaling — the moment agent-capacity metering becomes directly comparable to Exa's per-run and Parallel's per-request pricing. Vote its priority on the topic page.
Europe is excluded — and local-first architecture is suddenly a geographic moat
Cloud-tied host agents are geo-fenced, and that creates a live segmentation axis. "Dots will not reach Pro users in the European Economic Area, Switzerland or the UK for now" (Trending Topics), with GDPR/AI Act/DSA/DMA the suspected drivers; Meta's Muse agent is likewise US-only. The piece draws the contrast with OpenClaw directly: it "runs on the user's own hardware, so users keep control of their data. Dots, by contrast, run entirely in OpenAI's cloud and are tied to a subscription" — meaning "anyone who wants to use an agent modeled on Mr. Steinberger's idea right now has to turn to the original, because OpenClaw can be installed anywhere" (openai-deep-research-api-and-pro-tier).
This matters because it converts local-first from an ideology pitch into the only always-on-agent architecture available in the EEA/UK today. A model-agnostic orchestrator running on user hardware isn't competing with dots in Europe — it's the alternative to not having them at all.
What to watch: Any dots or Muse expansion into EEA/UK — that's the moment the compliance-driven local-first window closes. Steer the watch on the topic page.
The host is invading vertical distribution, too
Specialist dots get "their own identity, credentials, and access to the systems it needs" — IT-provisioned hardware and "deep integrations with a company's systems of record" — with Microsoft Agent 365 as the governance plane (OpenAI). That's OpenAI selling the same embed-into-the-system-of-record distribution that won for vertical players like OpenEvidence in the EHR — now from the host side, which squeezes the distribution moat that vertical research tools were counting on (openai-deep-research-api-and-pro-tier). If the host can provision credentials and sit in the system of record, "we're embedded where the work happens" stops being a defensible wedge on its own.
What to watch: Whether specialist dots ship with pre-built integrations for specific verticals (health, legal, finance) — that's the signal the host is eating vertical research distribution rather than just enabling it.
What surprised us
- OpenAI adopted the read-only research firewall as a built-in guardrail. After last cycle's incident-driven focus on agent governance, "proactive research" being restricted to read-only app access is notable — the trust axis is showing up as product architecture at the host, not just as compliance pressure (openai-deep-research-api-and-pro-tier
).
- The pricing move is aggressive in both directions at once. A $500 tier plus a halved $200 allowance, with the head of product arguing it "will net out at half the dollar in API spend compared to the old Pro $200 plan" (The New Stack) — that's not tiering up, that's repricing the ladder.
- Europe as the default home for agent alternatives is the unexpected market-map conclusion. The EEA/UK exclusion means the most regulated market is currently the least competitive one for always-on agents — an inversion of the usual dynamic where the US gets everything first.
Open threads worth a vote
- [watch] OpenAI publishes metered multi-dot pricing — fires when agent-capacity list prices land, resolving what the $500 plan actually includes. Prioritize it on the topic page.
- [watch] Dots or Muse launches in EEA/UK — fires when the local-first agent window closes. Prioritize it on the topic page.