{"slug":"deep-research","label":"Research Deep","item_count":3,"day_count":3,"source_count":3,"first_seen":"2026-06-15T13:56:33+00:00","last_updated":"2026-06-18T11:47:13+00:00","generated_at":"2026-07-06T10:03:08.720720+00:00","sources":["arxiv_cs_ai","arxiv_llm_reliability","aws_ml_blog"],"days":[{"date":"2026-06-15","items":[{"title":"Build context-rich research agents with Deep Agents and Bedrock AgentCore","url":"https://aws.amazon.com/blogs/machine-learning/build-context-rich-research-agents-with-deep-agents-and-bedrock-agentcore","source":"aws_ml_blog","type":"news","summary_1line":"In this post, you'll build a competitive research agent that demonstrates this pattern end to end. This walkthrough targets developers building multi-step AI workflows who need isolated execution environments for thei...","why_it_matters":"Matches feed focus: agent.","sid":"2f2101adc8737656","published":"2026-06-15T13:56:33+00:00","editor_note":"Anchors the thread in construction — an end-to-end build of a competitive research agent on Deep Agents and Bedrock AgentCore with isolated execution for multi-step workflows."}]},{"date":"2026-06-16","items":[{"title":"DRFLOW: A Deep Research Benchmark for Personalized Workflow Prediction","url":"http://arxiv.org/abs/2606.18191v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Deep research (DR) systems are increasingly used for complex information-seeking tasks, but existing works mainly focus on generating reports and summaries. In contrast, many enterprise tasks instead require an agent...","why_it_matters":"Matches feed focus: agent, eval.","sid":"0221a891a5b9e026","published":"2026-06-16T17:22:07+00:00","editor_note":"Redefines the goal toward enterprise needs: DRFLOW benchmarks personalized workflow prediction rather than report or summary generation."}]},{"date":"2026-06-18","items":[{"title":"ScaffoldAgent: Utility-Guided Dynamic Outline Optimization for Open-Ended Deep Research","url":"http://arxiv.org/abs/2606.20122v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Open-ended deep research (OEDR) requires systems to acquire knowledge through multi-round retrieval and generate coherent long-form reports. The outline plays a central role as a structural scaffold that coordinates r...","why_it_matters":"Matches feed focus: agent, eval.","sid":"b453c7e259c052f6","published":"2026-06-18T11:47:13+00:00","editor_note":"Turns to the report itself: ScaffoldAgent treats the outline as the optimization target, using utility-guided dynamic outline optimization for open-ended, multi-round-retrieval reports."}]}],"editorial":{"tldr":"In mid-June the target for deep research agents kept moving. AWS shipped a build pattern — an end-to-end competitive research agent on Deep Agents and Bedrock AgentCore with isolated multi-step execution. Then DRFLOW argued the real enterprise goal is predicting a personalized workflow, not generating a report, shifting what the agent is even supposed to optimize.","stale":false,"whats_new":"The newest item, ScaffoldAgent (Jun 18), moves the focus back to generation: it frames open-ended deep research as multi-round retrieval feeding a coherent long-form report and makes the outline the optimization target via utility-guided, dynamically optimized scaffolding.","why_it_matters":"What a deep research agent should be graded on is still unsettled — personalized workflow prediction versus outline-driven long-form coherence pull the design in different directions — and the criterion you pick changes what you build; AWS's Deep Agents + Bedrock AgentCore walkthrough is a ready reference for the multi-step plumbing under any of them.","take_for_builders":"If you're building a deep research agent, decide up front which bar you're optimizing — personalized workflow prediction or outline-driven long-form coherence — because they pull the design in different directions; the AWS Deep Agents + Bedrock AgentCore pattern gives you the isolated, multi-step execution plumbing regardless.","status":{"state":"Active research","tone":"rising","changed":"2026-06-18","detail":"A fast-moving research thread where the success criterion for deep research agents keeps being redefined — from build patterns to workflow prediction to report-generation structure."},"beats":[{"kicker":"BUILD PATTERN","tone":"launch","headline":"AWS ships an end-to-end research agent on Deep Agents + Bedrock AgentCore","summary":"A concrete walkthrough builds a competitive research agent with isolated execution for multi-step workflows.","sids":["2f2101adc8737656"]},{"kicker":"NEW TARGET","tone":"turn","headline":"DRFLOW benchmarks personalized workflow prediction, not report generation","summary":"Argues enterprise tasks need the agent to predict a personalized workflow rather than summarize.","sids":["0221a891a5b9e026"]},{"kicker":"NOW","tone":"now","headline":"ScaffoldAgent makes the outline the lever for long-form deep research","summary":"Utility-guided dynamic outline optimization for open-ended, multi-round-retrieval reports.","sids":["b453c7e259c052f6"]}],"open_questions":["Do these competing criteria — personalized workflow prediction versus outline-driven report quality — converge into one benchmark, or stay fragmented?","Does outline-first generation (ScaffoldAgent) measurably beat report-first approaches on open-ended deep research?","Will production stacks like Deep Agents + Bedrock AgentCore adopt any of these academic evaluation criteria?"],"generated_at":"2026-06-30T15:35:14Z"}}