Building Deep Research: How we Achieved State of the Art

Summary

Building production AI research agents presents challenges around context window pollution, escalating token costs, and architectural brittleness across model updates. To resolve these issues, Tavily rebuilt its deep research system around simplified orchestration, compact tooling, and active context curation. The agent mimics human research workflows by distilling tool outputs into concise reflections for ongoing reasoning, withholding raw retrieved web data until the final deliverable stage. Compared to traditional ReAct propagation architectures where token consumption scales quadratically, this reflection-based approach achieves linear token growth. As a result, the system reduced token consumption by 66% compared to Open Deep Research while reaching state-of-the-art performance on DeepResearch Bench.

Context

Long-horizon AI research agents face critical failure modes due to context window saturation, tool selection errors, non-deterministic drift, and quadratic token accumulation across iterative retrieval loops.

Approach / What changed

Tavily redesigned its deep research agent by simplifying orchestration logic, utilizing a minimal toolset, deduplicating search sources, and performing tool-side context curation via Tavily Advanced Search. Instead of propagating cumulative raw outputs across tool-calling iterations, the agent stores distilled reflections in short-term context and supplies raw content only when generating the final deliverable.

Takeaways

• Distilling tool outputs into reflections rather than propagating raw web tokens changes tool-calling context growth from quadratic to linear, reducing token consumption by 66% relative to Open Deep Research.
• Global state persistence and source deduplication keep agent context fresh, signal when information scope is narrowing to trigger broader exploration, and enable accurate source attribution.
• Providing a smaller, essential toolset reduces failure modes compared to large toolsets, helping models reliably choose tools and avoid iterative breakdowns.

Related reading

• Amazon · Jun 8, 2026

  Bridging intent and execution in agentic systems

  AI agent performance increasingly depends on the harness mediating interactions between language models and tools rather than raw model reasoning. To address mismatches between model intent and harness execution, the authors developed Simple Strands Agent, an open-source single-agent harness. The architecture implements robust tool-interface practices, such as condensing middle log outputs from bash commands, enforcing line-aware string replacements with ambiguity checks, and returning diff files after edits for verification. Evaluated across benchmarks like SWE-Bench-Verified, SWE-Bench-Pro, and Terminal-Bench 2, these model-harness alignments yielded consistent performance gains across frontier models without task-specific tuning. The findings show that model families require tailored interface codesign alongside stable evaluation infrastructure to achieve optimal benchmark reliability.

  Gaurav Gupta, Vatshank Chaturvedi

• Amazon · Aug 21, 2026