Cycle on Sep 8, 2026

I will begin by listing the open threads for this topic to understand what specific research questions or watches have been registered by previous cycles. This will help guide my exploration of the market for autonomous and AI research tools.

I will now read the existing notes for this topic to understand what has already been documented and see how we can expand upon or update our findings.

We have a rich set of existing notes! Some of the slugs we have are: - `market-map-positioning-hey-lefty` (Market Map & Positioning: Where Does Hey, Lefty Fit?) - `academic-ai-research-scholar-vs-synthesis` (Academic & Scientific AI Research: Scholar Labs and the Deep Search Paradigm) - `sacra-premium-private-market-mcp-data` (Sacra: Premium Private Market Research as Agent-Ready Infrastructure) - `perplexity-deep-research-consumer-agent` (Perplexity Deep Research: Search as Code, FactSet & Crunchbase Integrations, and the GEO Slop Defense) - `gemini-deep-research-agent` (Gemini Deep Research: Multi-Step Reasoning via the Interactions API) - `openai-deep-research-api-and-pro-tier` (OpenAI Deep Research: GPT-6 Astra, Recurrent Depth Architecture, and Daybreak Blue Gating) - `integrating-custom-mcp-servers` (Integrating Custom MCP Servers into Hey, Lefty's Research Cycles) Wait, the user's kickoff message contains several stories from Hacker News. Let's look at Theme 1: - `Prompting Is Dead in 6 Months. Andrew Ng, Stanford [video]` — https://news.ycombinator.com/item?id=49605222 Let's search Hacker News or use `hn_comments` to read the discussion for this story and other relevant stories. Wait, let's look at the story list on HN and see if there are other stories that are highly relevant to our topic. Let's search for "Andrew Ng" or "Prompting Is Dead" or "Gemini Deep Research" or "Perplexity" on Hacker News, or let's use `hn_comments` on the Andrew Ng video item: `49605222`. Let's check `hn_comments` on `49605222` to see what Andrew Ng is saying about prompting and agentic loops, which directly relates to the underlying technology of AI research tools (the shift from prompting to agentic loops). This is highly relevant to our topic of autonomous/AI research tools and their core technology! Let's do that first.

Ah! The Andrew Ng video `Prompting Is Dead in 6 Months` is a mislabeled upload of stolen content from a 2025 career advice lecture, and the description claiming "Graphs are what's replacing it" is a total misrepresentation. This is a fascinating meta-finding about AI content farming and hype cycles, but let's focus on the actual market for autonomous/AI research tools requested by the user. Let's do a web search to gather information on Gemini Deep Research, Perplexity Deep Research, Sacra, and other AI research tools, their features, value props, core technologies, data sources, pricing, and business models. Let's search for: 1. "Gemini Deep Research" features pricing technology 2. "Perplexity Deep Research" features pricing technology 3. "Sacra" research platform features pricing business model 4. "Scholar Labs" AI research or "Deep Search" academic tools 5. "AI research tools" market map or comparison Let's start searching.

Wow, this is incredibly valuable. We see: - Gemini Deep Research is accessible via the **Interactions API** (which is now Generally Available as of June 2026). - The models used are `deep-research-preview-04-2026` and `deep-research-pro-preview-12-2025` (powered by Gemini 3 Pro). Wait, the article mentions `deep-research-pro-preview-12-2025` and Gemini 3 models (Gemini 3 Flash, Gemini 3 Pro). - The Interactions API supports server-side state management (`previous_interaction_id`), background execution (`background=true`), and observable execution steps. - Pricing structure: Billing comprises model usage (tokens) and tool execution (search and grounding). They use an `is_deep_research` billing label. There is a deferred tier for background jobs. - Key features: multi-phase workflows (planning, task execution, synthesis). Let's search for "Sacra" to understand its research platform. What does Sacra do? What is its business model? Where does its data come from? Who do they sell to? Let's do a web search for "Sacra" research platform.

Let's search more deeply on Sacra's data, API, and MCP integration. Our prior notes mentioned: "Sacra has established itself as the leading provider of 'agent-ready infrastructure' for private market intelligence. By unbundling its high-quality, proprietary financial datasets and analyst..." Wait! Let's read the note `sacra-premium-private-market-mcp-data` to get the details we already have on Sacra. Let's read `sacra-premium-private-market-mcp-data` and `perplexity-deep-research-consumer-agent` and `market-map-positioning-hey-lefty`! This is very important. I will read them. Oh wait, I don't have a read_note tool that reads a specific note's body, but I can use `write_note` to update. No, wait, I can read notes via `read_notes` which returns previews, but wait, can I get the full text of a note? Oh, wait, the prompt says "The topic you've researched before already has standing findings — the 'Relevant prior findings' block, plus `read_notes` for the rest." Ah, let's see if there's a tool to read the full body of a note. There isn't a direct "read_note_body" tool, but we can search for the notes or we can fetch them? No, we don't have a specific tool to read the full body of our own notes, but we have their previews in `read_notes` and some are in the kickoff message's "Relevant prior findings" block. Let's see if we can search the web for "Sacra MCP server" or "Sacra agent-ready infrastructure" to get more details on how Sacra integrates with AI agents. Let's do that.

This is incredibly detailed! Let's fetch the actual pages of: 1. `https://sacra.com/p/introducing-sacra-mcp-connector-chatgpt/` 2. `https://sacra.com/p/introducing-sacra-connector-for-claude/` 3. `https://stytch.com/customer-stories/sacra` Let's fetch `https://sacra.com/p/introducing-sacra-connector-for-claude/` first.

Let's fetch `https://stytch.com/customer-stories/sacra` to understand how Sacra handles authentication for agents and what its technology stack looks like. This is highly relevant to our findings about agent-ready infrastructure and EMA!

This is an incredible case study of Sacra's technical architecture and how it utilizes Stytch for predictable agent authentication (MCP / OAuth / Connected Apps). Let's do a search on Scholar Labs and other academic/literature AI research tools (e.g., Google Scholar, ScholarAI, Consensus, Elicit, etc.) to understand their features, core technology, data sources, pricing, and business models. Let's search for "Scholar Labs" AI research or "Scholar Labs" "Deep Search". Let's do a web search.

This is highly relevant! I should fetch `https://aarontay.substack.com/p/scholar-labs-early-review-google` to get Aaron Tay's framework of "Deep Search" vs "Deep Research" in academic/literature research tools. This represents a core structural division in the market map. Let's fetch that page.

This is an outstanding, deep, analytical source. Let's summarize the key points from Aaron Tay's piece: 1. **Deep Search vs. Deep Research**: - **Deep Search** (e.g., Scholar Labs, AI2 Paper Finder / ASTA): Focuses on finding relevant papers rather than generating direct answers. It spends extra time/compute on iterative search strategies and relevance reasoning (assessing paper-by-paper and generating a "rationale" for relevance). It prioritizes precision in ranking. It *does not* attempt to synthesize across multiple papers. - **Deep Research** (e.g., Gemini Deep Research, OpenAI Deep Research, Scopus Deep Research): Focuses on generating a long-form report, synthesis, and visualizations across multiple papers. This is a much trickier task because it requires balancing conflicting findings across different contexts. 2. **Google Scholar Labs (Scholar Labs)**: - Launched: November 18, 2025. - Core Value Prop: Leverages Google Scholar's massive, unparalleled index (which dwarfs OpenAlex and Semantic Scholar and includes full-text from major publishers). - Core Technology: Analyzes the natural language question to identify key topics, aspects, and relationships, searches them on Scholar, evaluates the results paper-by-paper using an LLM to generate a rationale/explanation for why each paper answers the question, and displays them. It is conservative and does not synthesize across papers. - Target Audience: Academic researchers, students, and professionals searching for peer-reviewed literature. - Pricing/Business Model: Free/included in Google Scholar (currently in Labs/waitlist). Let's do a web search on other tools mentioned: - Perplexity Deep Research (we have a note, but let's see if there is more recent info on its pricing, business model, and Search as Code architecture). - Sacra (already covered nicely: premium private market data, $50-$1,500/mo, MCP connector for Claude & ChatGPT powered by Stytch Connected Apps). - OpenAI Deep Research (Astra, recurrent depth architecture, Daybreak Blue gating). Let's search for "OpenAI Deep Research" features and pricing. - What about other financial/business AI research tools like AlphaSense, Tegus, or others? Let's check. Let's search for: 1. "OpenAI Deep Research" pricing features 2. "Perplexity Deep Research" pricing features 3. "AlphaSense" AI features pricing 4. "Tegus" AI features pricing Let's search.

This is gold! We have solid, current, and highly specific data from 2026. Let's analyze what we have here. Let's do a quick search on "Perplexity Deep Research" or "Perplexity Pro" pricing and features in 2026 to see how it fits in. Let's search.

This is phenomenal. We have extremely detailed and structured data for: 1. **OpenAI Deep Research (ChatGPT & API)**: - ChatGPT Plus ($20/mo) gets 10 runs/month; Pro ($200/mo) gets 250 runs/month. - Powered by o3-deep-research (and historically o4-mini-deep-research, though o4-mini is deprecated). - o3-deep-research is priced at $10/M input, $40/M output, plus $10 per 1,000 web search calls. - Supports MCP servers, code interpreter, and web search. - Background Mode (GPT-5.5 / GPT-6 Astra) allows long-running jobs to run without keeping the app in the foreground. 2. **Google Gemini Deep Research**: - Models: `deep-research-preview-04-2026` and `deep-research-max-preview-04-2026`. - Access: Interactions API, GA as of June 2026. - Features: Multi-step reasoning, google search, url context, code execution, MCP, file search, visualization, multimodal. - Pricing: per-token + per-search (~$1-3 per task for standard, ~$3-7 per task for Max). Max runs up to 160 searches and up to 60 min. 3. **Perplexity Deep Research**: - Models: `sonar-deep-research` (endpoints retiring Sep 27, 2026, transitioning to Agent API preset "high"). - Pricing: $2/M input, $8/M output, plus $2/M citation tokens, $3/M reasoning tokens, and $5 per 1,000 searches. - Pro tier ($20/mo) gets 20 Deep Research queries/day. Max tier ($200/mo) gets unlimited Deep Research queries. - Enterprise Pro ($40/seat/mo) and Enterprise Max ($325/seat/mo). - Core Architecture: **Search as Code (SaC)**. The model writes code that compiles search itself, running thousands of retrieval steps in parallel. 4. **Sacra**: - Focuses on private market intelligence (pre-IPO, growth-stage, early-stage companies). - Pricing: $50/mo Standard, $350/mo Pro, $1,500/mo Platform. - Features: High-quality, proprietary financial datasets, analyst reports, news, funding rounds. - Integrates with AI agents (Claude, ChatGPT) via the **Model Context Protocol (MCP)**, built on Stytch Connected Apps for secure and stable B2B/consumer authentication. Wait, let's look at the open threads we listed: - `OpenAI Releases Public Post-Mortem on Hugging Face and Wiki Collusion Incidents` (fire_date: 2026-10-01) - Current date is 2026-09-08, so this hasn't fired yet. - `Google Resolves Gemini Deep Research Max MCP Invocation Regression` - Let's see if this has been resolved or if we can find any information about it. - `Anthropic Launches Scientific or Biological AI Model under John Jumper` - Let's see if there is any news on this. - `Antitrust Scrutiny of Anthropic Silent Safeguards` - `Appeals and Rulings on AI Summary Liability in the EU` - `Gemini Deep Research Max Transitions to GA` - Wait, our search results say: "The Interactions API is the best way to build with Gemini models and agents. As of June 2026, it is Generally Available and recommended for all new projects." But wait, what about the models `deep-research-preview-04-2026` and `deep-research-max-preview-04-2026`? Are they GA or still in preview? Let's check. Let's search the web for: - "Gemini Deep Research" "GA" or "general availability" - "Gemini Deep Research Max" "MCP" - "Anthropic" "John Jumper" model launch - "Munich Regional Court" "AI" summary appeal or "Case 26 O 869/26" Let's do a search on these.

This is incredibly precise pricing and performance data! Let's analyze this comparison between Gemini Deep Research and OpenAI Deep Research. - Gemini Deep Research uses standard Gemini 3.1 Pro inference rates: $2.00 per 1M input tokens, $12.00 per 1M output tokens (including thinking/reasoning tokens). Wait, the table shows input: $2.00, output: $12.00. - Tool usage: Google Search Grounding is billed per existing pricing (e.g., $14 per 1k searches). Gemini Deep Research runs about 80 searches per task (~$1.12), while Gemini Deep Research Max runs about 160 searches per task (~$2.24). - A typical task costs ~$2.00 for standard Gemini Deep Research, and ~$5.00 for Gemini Deep Research Max. - OpenAI's o3-deep-research costs $10/M input, $40/M output, plus $10 per 1k search calls (typically 15-20 searches, though o3-deep-research can do more). A typical o3-deep-research query can cost ~$1-10 depending on reasoning depth. Let's check if there are any other open threads. Wait! One of the open threads is: "Monitor when Google DeepMind officially resolves the remote MCP tool invocation regression in Gemini Deep Research Max (deep-research-max-preview-04-2026) where the agent fails to send tools/call requests after tool discovery." Let's search for "Gemini Deep Research Max" "MCP" regression or "tools/call" or "fails to send" or "Interactions API" MCP. Let's do a search.

This is incredibly exciting and directly addresses the open thread: - `Google Resolves Gemini Deep Research Max MCP Invocation Regression` Wait! The forum post `https://discuss.ai.google.dev/t/regression-gemini-deep-research-max-generates-an-ungrounded-report-after-discovering-but-never-invoking-mcp-tools/180801` is on the Google AI Developers Forum. It states: "The primary production regression affects `deep-research-max-preview-04-2026`. Gemini connects to our remote MCP server, completes initialization, and retrieves the tool catalog, but never sends a `tools/call` request." "The Gemini integration successfully invoked the same MCP tools earlier on September 2, 2026. A later run that day stopped invoking them." Wait, has Google resolved this regression? Let's check if there are any updates on that thread or other threads. Let's do a search on the forum for "Regression: Gemini Deep Research Max" or "180801" to see if there are any replies or resolutions. Let's do that.

Ah, the thread has no replies on Google AI Developer Community Forum yet. So the regression is still unresolved as of our current date (September 8, 2026) — it was reported around September 2, 2026, and remains open. This is a very valuable and current piece of technical information. I will note this! Let's do a search on Anthropic's John Jumper model or science features to see if there's any update. Let's do a web search.

This is absolutely incredible! The web search budget is spent, but we have *all* the evidence we need to write rich, detailed, grounded findings. Let's look at what we've discovered: 1. **Anthropic's Claude Science**: - Launched: June 30, 2026. - What it is: A standalone flagship AI workbench for scientists (specifically focused on biology, life sciences, and chemistry). - Core Value Prop: Sparing scientists from bouncing between databases, pipelines, and tools. Act as a "scientific operating environment" rather than a chatbot. - Core Technology: Integrates over 60 scientific databases (like UniProt, PDB, Ensembl, Reactome, ChEMBL, GEO), prebuilt toolkits, and local/remote execution (Python, R, shell commands). It uses a "reviewer agent" to check claims against the execution record and produces auditable artifacts. - Models: It does *not* use a new specialized model; it runs existing Claude models (Claude Opus 4.8, Claude Mythos Preview) available to everyone. - Leadership: John Jumper (Nobel laureate, AlphaFold co-creator) left Google DeepMind to join Anthropic just 11 days before the Claude Science launch! - Pricing/Business Model: Available to all paid Claude subscribers (part of their existing plans), with a beta program offering research-lab discounts. 2. **OpenAI's Deep Research and GPT-Rosalind**: - OpenAI's Deep Research is powered by `o3-deep-research`. Plus tier ($20/mo) gets 10 queries/mo; Pro ($200/mo) gets 250 queries/mo. - API pricing: $10/M input, $40/M output, plus $10 per 1,000 web searches. - GPT-Rosalind: Released in April 2026 as a specialized model fine-tuned for biological reasoning, limited to qualified enterprise customers in the US. 3. **Google Gemini Deep Research & Scholar Labs**: - Gemini Deep Research (standard & Max) available via GA Interactions API as of June 2026. - Standard typical cost: ~$2.00 per task (runs ~80 searches). Max typical cost: ~$5.00 per task (runs ~160 searches). - Scholar Labs: Launched quietly on November 18, 2025. It is a "Deep Search" tool, not "Deep Research". It focuses on paper-by-paper evaluation and relevance reasoning rather than synthesis across papers. Free/included in Google Scholar. - MCP Regression: A major production regression reported on September 2, 2026, where `deep-research-max-preview-04-2026` and `deep-research-preview-04-2026` connect to remote MCP servers, retrieve the tool catalog, but never send a `tools/call` request (and yet generate ungrounded reports representing private data as retrieved). This remains unresolved as of September 8, 2026. 4. **Perplexity Deep Research**: - Sonar Deep Research API pricing: $2/M input, $8/M output, plus $2/M citation tokens, $3/M reasoning tokens, and $5 per 1,000 searches. Average cost is $0.41–$1.32 per query. - Individual pricing: Pro ($20/mo) gets 20 Deep Research queries/day; Max ($200/mo) gets unlimited Deep Research queries. - Enterprise: Pro ($40/seat/mo), Max ($325/seat/mo). - Core Technology: **Search as Code (SaC)**. The model writes Python code that programmatically coordinates search, running thousands of retrieval steps in parallel. 5. **Sacra**: - Focuses on private market intelligence. - Pricing: $50/mo Standard, $350/mo Pro, $1,500/mo Platform. - Features: High-quality, proprietary financial datasets and analyst reports. - Core Technology: Uses Model Context Protocol (MCP) to connect directly to Claude/ChatGPT. Uses Stytch Connected Apps for secure, predictable B2B/consumer authentication and dynamic client registration. Let's look at how we should update our findings and resolve/record threads and watches. First, let's check the open threads and see if we can resolve any of them: - `Anthropic Launches Scientific or Biological AI Model under John Jumper` (ID: `019ee351-7287-781b-9912-104ad72fc9bf`). This watch was for Anthropic launching a scientific or biological AI model or feature under the leadership of John Jumper. Claude Science was launched on June 30, 2026, and John Jumper left DeepMind for Anthropic 11 days before the launch! This watch has fired! Let's record an event and resolve it. Wait, let's see if we should call `record_event` on `019ee351-7287-781b-9912-104ad72fc9bf`. Yes! Let's call `record_event` with `resolves: true` because it's a `one_shot` watch. Let's check the exact details. The watch title is "Anthropic Launches Scientific or Biological AI Model under John Jumper". Wait! Let's also check: - `Google Resolves Gemini Deep Research Max MCP Invocation Regression` (ID: `01a07500-f5e5-76cc-b857-7c3cca66f868`). This watch is for Google DeepMind resolving the remote MCP tool invocation regression. Our search showed that as of September 8, 2026, this regression is *still unresolved* (it was posted on September 2, 2026, and remains open with no replies on the Google AI Developers Forum). So we do not resolve this watch yet. - `Gemini Deep Research Max Transitions to GA` (ID: `019ea05f-d685-7208-a431-55568fba4987`). The Interactions API is GA as of June 2026, but the models `deep-research-preview-04-2026` and `deep-research-max-preview-04-2026` are still listed as preview models in the docs (and the forum post from September 2, 2026, refers to them as preview models). So this watch remains open. Let's call `record_event` on the Anthropic scientific model watch. Let's do that.

Now, let's write our findings! Let's see: we have existing notes that we can update or expand. The existing notes are: - `market-map-positioning-hey-lefty` (Market Map & Positioning: Where Does Hey, Lefty Fit?) - `academic-ai-research-scholar-vs-synthesis` (Academic & Scientific AI Research: Scholar Labs and the Deep Search Paradigm) - `sacra-premium-private-market-mcp-data` (Sacra: Premium Private Market Research as Agent-Ready Infrastructure) - `perplexity-deep-research-consumer-agent` (Perplexity Deep Research: Search as Code, FactSet & Crunchbase Integrations, and the GEO Slop Defense) - `gemini-deep-research-agent` (Gemini Deep Research: Multi-Step Reasoning via the Interactions API) - `openai-deep-research-api-and-pro-tier` (OpenAI Deep Research: GPT-6 Astra, Recurrent Depth Architecture, and Daybreak Blue Gating) - `integrating-custom-mcp-servers` (Integrating Custom MCP Servers into Hey, Lefty's Research Cycles) Wait, the prompt says: "Your findings are a LIVING SET, not a fresh page. A topic you've researched before already has standing findings — the 'Relevant prior findings' block, plus `read_notes` for the rest. This cycle CONTINUES that set: you revise and extend it as the story develops, you do not start over and re-mint everything under new names." "Name a slug for the DURABLE SUBJECT it tracks... When this cycle advances a subject you already have a finding for, call `write_note` with that SAME slug to revise it in place — an upsert that snapshots a new revision. Mint a fresh slug ONLY for a genuinely new subject." Let's review the notes we want to update. 1. `market-map-positioning-hey-lefty`: This is the central market map and positioning note. We should update it with our new, comprehensive market map that places Gemini Deep Research, Scholar Labs, Perplexity Deep Research, OpenAI Deep Research, Claude Science, Sacra, and Hey, Lefty into a clear, multi-dimensional matrix. 2. `academic-ai-research-scholar-vs-synthesis`: We can update this with Google's Scholar Labs (launched Nov 18, 2025) and Anthropic's Claude Science (launched June 30, 2026), contrasting Scholar Labs' "Deep Search" paradigm with Claude Science's "scientific operating environment" workflow paradigm (and OpenAI's GPT-Rosalind). 3. `sacra-premium-private-market-mcp-data`: We can update this with the technical details of Sacra's Stytch Connected Apps authentication architecture, which enables seamless B2B/consumer transition and stable MCP/OAuth dynamic client registration. 4. `perplexity-deep-research-consumer-agent`: We can update this with Perplexity's new pricing models (Pro $20/mo, Max $200/mo, Enterprise Max $325/seat/mo), the Sonar Deep Research API pricing structure (billing citation, reasoning, and search tokens), and its Search as Code (SaC) architecture. 5. `gemini-deep-research-agent`: We can update this with the GA release of the Interactions API (June 2026), its pricing comparison with OpenAI (~$2.00 vs ~$1.12 search cost per task), and the active MCP tool invocation regression that was reported on September 2, 2026, and remains unresolved. This is a perfect, logical, highly-structured plan. Let's write these notes one by one. Let's start with `academic-ai-research-scholar-vs-synthesis`. Wait, let's verify what the prompt requires for body shape: "No YAML frontmatter — the title is the first `# ` heading and the slug is the slug parameter you pass." "Every substantive finding must: - name and link the actual thing (the story, article, paper, or page that was posted/published) — not just "the discussion", - carry 1-2 REPRESENTATIVE VERBATIM QUOTES, copied exactly, each kept with the link it came from (an HN comment permalink, the article URL, etc.), - then say what it means." Let's carefully construct the markdown for `academic-ai-research-scholar-vs-synthesis`. Wait, let's look at the quotes we collected: - From Aaron Tay's substack: "The main differentiator is that a Deep Search tool is designed to find relevant papers and not generate direct answers. It differs from normal search by going “deeper,” running beyond the typical 1000ms latency you expect from Google or conventional databases." (https://aarontay.substack.com/p/scholar-labs-early-review-google) - From Anthropic's Claude Science launch: "Claude Science is an app that integrates the tools and packages that researchers most commonly use, produces auditable artifacts, and provides flexible access to computing resources." (https://www.anthropic.com/news/claude-science-ai-workbench) - From TechCrunch: "To be clear, Anthropic says Claude Science is 'not a new AI model and not a more capable model for biology. It runs the same Claude models already available to everyone today (including Claude Opus 4.8), with no special access and no gating.'" (https://techcrunch.com/2026/06/30/anthropics-claude-science-bets-on-workflow-not-a-new-model-to-win-over-scientists/) Let's write the updated note for `academic-ai-research-scholar-vs-synthesis`.

Now, let's update `sacra-premium-private-market-mcp-data` to reflect the technical integration of Sacra's agent-ready infrastructure, specifically focusing on its Stytch-powered Connected Apps auth, magic link login unification, and dynamic client registration for Claude and ChatGPT MCP servers. Let's look at the quotes we collected: - From Sacra's Stytch case study: "MCP is the handshake between our data and agent workflows." (https://stytch.com/customer-stories/sacra) - From Sacra's Stytch case study: "Stytch’s MCP-ready flows made our ChatGPT and Claude connectors predictable: dynamic client registration, safe resets, and clear errors we can action instead of mystery failures." (https://stytch.com/customer-stories/sacra) - From Sacra's Claude MCP announcement: "If you research private companies with AI today, you mostly get SEO spam. Marketers create endless pages about companies—often without any real data—just to rank on search... That's why we built Sacra MCP." (https://sacra.com/p/introducing-sacra-connector-for-claude/) Let's write the updated note for `sacra-premium-private-market-mcp-data`.

Now, let's update `perplexity-deep-research-consumer-agent` to reflect Perplexity's 2026 pricing model (Pro $20/mo, Max $200/mo, Enterprise Max $325/seat/mo), the Sonar Deep Research API pricing structure (billing citation, reasoning, and search tokens), and its Search as Code (SaC) architecture. Let's look at the quotes we collected: - From Perplexity blog: "Deep Research in Computer is built on our Agent Search SDK and Search as Code architecture. We let the model write code that assembles search itself, running thousands of retrieval steps, in parallel, tailored to each question." (https://www.perplexity.ai/hub/products/deep-research) - From Suprmind pricing analysis: "sonar-deep-research has the most distinctive billing structure in the Sonar API. Beyond standard input and output tokens, it charges separately for citation tokens ($2/M), reasoning tokens ($3/M), and search queries ($5/K)." (https://suprmind.ai/hub/perplexity/pricing/) - From CloudZero API pricing: "You don’t directly control how many searches run. The model decides. A single query can trigger dozens of searches, which makes costs less predictable... Total: ~$0.30 to $1.30+ per query, depending on context depth." (https://www.cloudzero.com/blog/perplexity-api-pricing/) Let's write the updated note for `perplexity-deep-research-consumer-agent`.

Now let's update `gemini-deep-research-agent` to reflect the GA release of the Interactions API (June 2026), its pricing structure, typical cost per task (~$2.00 vs ~$1.12 search cost), and the active, unresolved MCP tool invocation regression reported on September 2, 2026. Let's look at the quotes we collected: - From the Google AI Developer Community Forum regression post: "The primary production regression affects `deep-research-max-preview-04-2026`. Gemini connects to our remote MCP server, completes initialization, and retrieves the tool catalog, but never sends a `tools/call` request. It nevertheless generates a substantive report that refers to private source data without retrieving it." (https://discuss.ai.google.dev/t/regression-gemini-deep-research-max-generates-an-ungrounded-report-after-discovering-but-never-invoking-mcp-tools/180801) - From TokenCost analysis: "Where o4-mini-deep-research uses 15-20 queries per task at $10/1K, Gemini Deep Research uses 80 at $14/1K." (https://tokencost.app/blog/gemini-deep-research-agent-cost) - From Google's Interactions API documentation: "The Interactions API is the best way to build with Gemini models and agents. As of June 2026, it is Generally Available and recommended for all new projects." (https://ai.google.dev/gemini-api/docs/interactions-overview) Let's write the updated note for `gemini-deep-research-agent`.

Now, let's update `market-map-positioning-hey-lefty` to synthesize all of these insights into a definitive market map and positioning strategy for Hey, Lefty. Let's look at the positioning. Hey, Lefty is positioned as a "model-agnostic, multi-source research orchestrator." Let's see: how does this fit into the current landscape of: - General-purpose Deep Research (OpenAI Deep Research, Gemini Deep Research, Perplexity Deep Research) - Specialized Domain Search (Google Scholar Labs, AI2 Paper Finder) - Domain-specific Workflow Environments (Anthropic Claude Science, Sacra) Let's write this update. It will be incredibly rich, structured, and strategic. We will use the exact slug `market-map-positioning-hey-lefty`. Let's do that!

All five notes are beautifully updated and expanded with highly precise, current, and grounded evidence. Let's check if there are any other open threads or watches we should register. Let's see: we have an ongoing watch: - `Appeals and Rulings on AI Summary Liability in the EU` (ID: `019eafd3-1ce9-7376-80d4-c4f5e9580f9d`) - `Antitrust Scrutiny of Anthropic Silent Safeguards` (ID: `019eafd3-288d-7900-9b7b-f29c43234577`) - `OpenAI Releases Public Post-Mortem on Hugging Face and Wiki Collusion Incidents` (ID: `01a07a23-cada-7d9d-9bd4-5560d7f0b649`) - `Google Resolves Gemini Deep Research Max MCP Invocation Regression` (ID: `01a07500-f5e5-76cc-b857-7c3cca66f868`) - `Gemini Deep Research Max Transitions to GA` (ID: `019ea05f-d685-7208-a431-55568fba4987`) Wait, are there any new forward-looking developments that we should register as watches? Yes, let's look at: - Perplexity's Sonar API endpoint retirement on September 27, 2026, transitioning to the Agent API preset "high". This is a dated catalyst! Let's register a watch for this. - `cardinality`: `one_shot` - `fire_date`: `2026-09-27T00:00:00Z` - `watch_type`: `dated_catalyst` - `title`: `Perplexity Sonar Deep Research Endpoints Retire` - `description`: `Monitor when Perplexity officially retires its sonar-deep-research endpoints on September 27, 2026, and transitions API developers to the new Agent API preset "high".` Let's call `record_watch` to register this. This is extremely clean and matches our guidelines perfectly!

Excellent! Let's double check if we need to resolve any threads. Wait, let's see if we resolved the watch for Anthropic launching a scientific model under John Jumper. We recorded an event on thread `019ee351-7287-781b-9912-104ad72fc9bf`, which was a watch of kind `watch` (cardinality `one_shot`). Let's call `resolve_thread` on it to mark it as resolved by this cycle. Let's do that.

This is incredibly clean. I have updated the relevant prior findings, resolved the appropriate thread, recorded a new watch, and documented the entire market map and technical landscape of autonomous AI research tools. I will now call `finish_cycle` to complete this research cycle.

I have completed the research cycle for topic **019e8498-f497-7eb3-9d41-64bb48fe1e5d** (The Market for Autonomous and AI Research Tools). Here is a summary of the actions taken and findings generated during this cycle: ### 1. Active Thread & Watch Resolutions * **Resolved Watch:** *Anthropic Launches Scientific or Biological AI Model under John Jumper* (ID: `019ee351-7287-781b-9912-104ad72fc9bf`). Anthropic officially launched **Claude Science** on June 30, 2026—a major standalone AI workbench for computational biology and chemistry. This launch followed Nobel laureate John Jumper (co-creator of AlphaFold) leaving Google DeepMind to join Anthropic just 11 days prior. * **Registered New Watch:** *Perplexity Sonar Deep Research Endpoints Retire* (ID: `01a07f4b-8589-7ac8-892d-656abf8236e3`). Perplexity has scheduled the retirement of its `sonar-deep-research` API endpoints for **September 27, 2026**, transitioning developers to its new Agent API preset "high". * **Monitored Ongoing Issue:** *Google Resolves Gemini Deep Research Max MCP Invocation Regression* (ID: `01a07500-f5e5-76cc-b857-7c3cca66f868`). As of September 8, 2026, the critical regression reported on September 2, 2026 (where Gemini Deep Research Max connects to remote MCP servers and retrieves the catalog but fails to execute tool calls) remains unresolved. --- ### 2. Living Findings Updated & Expanded I have updated our core findings with highly specific, grounded 2026 market data to construct a robust strategic landscape for **Hey, Lefty**: #### 🗺️ **Market Map & Positioning: Where Does Hey, Lefty Fit?** (slug: `market-map-positioning-hey-lefty`) * **Synthesis:** Mapped the AI research landscape into four quadrants across two axes: *Scope of Inquiry* (General Web vs. Domain-Specific) and *Interaction Paradigm* (Search/Discovery vs. Workflow/Orchestration). * **Hey, Lefty Positioning:** Positioned Hey, Lefty as the premier **Model-Agnostic, Multi-Source Research Orchestrator**. Hey, Lefty occupies a high-value, unfilled niche by acting as a unified control plane that orchestrates multiple models (e.g., routing planning to Gemini, web grounding to Perplexity, and synthesis to Claude) and connects to arbitrary, private data sources via the Model Context Protocol (MCP). This model-agnostic approach also shields enterprise users from provider-specific bugs, such as Google's active MCP regression. #### 🧪 **Academic & Scientific AI Research: Scholar Labs and the Deep Search Paradigm** (slug: `academic-ai-research-scholar-vs-synthesis`) * **Key Findings:** Contrasted Google's **Scholar Labs** (launched November 18, 2025) with Anthropic's **Claude Science** (launched June 30, 2026). * **Scholar Labs:** Operates on a conservative "Deep Search" paradigm. It evaluates papers paper-by-paper using LLM relevance reasoning to generate custom "rationales," completely avoiding the high-hallucination risk of multi-paper synthesis. * **Claude Science:** Operates as a "Scientific Operating Environment" or workbench. It integrates over 60 scientific databases with local/remote code execution (Python, R, shell) and features a "reviewer agent" that automatically audits execution records to verify scientific claims. #### 📈 **Sacra: Premium Private Market Research as Agent-Ready Infrastructure** (slug: `sacra-premium-private-market-mcp-data`) * **Key Findings:** Documented Sacra's transition from a consumer-facing platform to an "agent-ready infrastructure" provider for private market intelligence, bypassing SEO-ridden public search. * **Technical Integration:** Detailed Sacra's authentication architecture powered by **Stytch Connected Apps**. This setup supports dynamic client registration (DCR) and predictable OAuth resets for Claude and ChatGPT MCP servers. It also enables robust paywall metering via device fingerprinting and allows Sacra to transition from consumer to B2B multi-tenant structures without touching its agentic MCP flows. #### 🔍 **Perplexity Deep Research: Search as Code, FactSet & Crunchbase Integrations, and the GEO Slop Defense** (slug: `perplexity-deep-research-consumer-agent`) * **Key Findings:** Detailed Perplexity's **Search as Code (SaC)** architecture, where the model programmatically writes and executes Python search scripts in parallel sandboxes to dynamically coordinate retrieval. * **Pricing Models:** Documented Perplexity's 2026 pricing tiers (Pro at $20/mo, Max at $200/mo, Enterprise Max at $325/seat/mo) and analyzed its complex **five-meter Sonar Deep Research API pricing structure** (billing input, output, citation, reasoning, and search query tokens), which drives single-query costs to $0.30–$1.32+ based on search depth and thinking time. #### 🤖 **Gemini Deep Research: Multi-Step Reasoning via the Interactions API** (slug: `gemini-deep-research-agent`) * **Key Findings:** Analyzed Google's GA **Interactions API** (released June 2026), which handles server-side state management (`previous_interaction_id`), background execution (`background=true`), and observable reasoning steps. * **Cost Dynamics:** Grounded typical task costs (standard Gemini Deep Research runs ~80 searches costing ~$2.00/task; Gemini Deep Research Max runs ~160 searches costing ~$5.00/task) and contrasted them with OpenAI's `o3-deep-research` API rates. * **MCP Regression:** Fully documented the mechanics of the active, unresolved September 2026 regression where Gemini Deep Research Max successfully retrieves MCP tool catalogs but fails to invoke tool calls, yet still generates ungrounded reports referencing private data. All findings are now live, fully updated, and cross-linked via stable wikilinks to ensure a cohesive, persistent, and highly actionable research artifact for your team.