--- title: "Slides: The 100-Tool Agent Is a Trap - Sohail Shaikh & Ankush Rastogi, Prosodica" category: "slides" video_id: "vh2VGuQ3zhY" sourceLabels: ["Public YouTube video frames", "Public YouTube metadata"] --- # Slides: The 100-Tool Agent Is a Trap - Sohail Shaikh & Ankush Rastogi, Prosodica ## Source Video [The 100-Tool Agent Is a Trap - Sohail Shaikh & Ankush Rastogi, Prosodica](https://www.youtube.com/watch?v=vh2VGuQ3zhY) ## Relationship To World's Fair 2026 These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material. ## Related Scheduled Sessions - No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck. ## Extracted Slides ![[assets/slides/vh2VGuQ3zhY/slide-007.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-007.html) - AI slide classifier: `content_slide` confidence `0.98` - Text source: advanced OCR `rapidocr-live/border-trim/contrast`. - OCR decision: ready — Dense multi-card slide with small bullet text and mixed layout. Slide text: > HOWTO BUILD IT > 3 Step Implementation Pattern > Ankush Rastogi > Build Tool Index Route Each Query Inject & Call LLM > OFFLINE·ONE-TIME RUNTIME·EVERYREQUEST RUNTIME·EVERYREQUEST > Collect all tool names, descriptions,and JsON schemas Embed the incominguser query with thesamemodel Fetch JsON schemas for the selected tools only > Index once and reuse forever by Embed each tool description SentenceTransformers) Pinecone. storingvectors in FAisS or (OpenAl,Cohere, Run approximate Retrieve top-Ktools（K=5 the toolindex nearest-neighbor search against default); apply a cosine threshold Build thepromptwith just those Call the LLM → return result; log selections for monitoring schemas in the toolsparameter Sohaa Shukh > Antush Rastogi ![[assets/slides/vh2VGuQ3zhY/slide-008.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-008.html) - AI slide classifier: `content_slide` confidence `0.99` - Text source: advanced OCR `rapidocr-live/border-trim/contrast`. - OCR decision: ready — Six-item checklist with smaller subtext across two columns. Slide text: > YOURROADMAP > Implementation Checklist > Ankush Rastogi > Catalog Your Tools Build the Embedding Index > Gather alltoolnames,descriptions,and JsON schemasina structured list Pinecone,one-time setup Embed each description, store vectors in FAiss or > Implement the Router Integrate Into the Agent Loop > schemas embed(query)→similarity_search→top-K→fetch Replace your static functions list with therouter output on every call Sohaa Shakh > 5 Benchmark accuracy at K=3,5,10on a held-out set; pickthe sweet spot Evaluate & Tune K 6 your catalog grows Monitor & Iterate Log selections;alert on missed tools;re-embed as > 14 > Antush Rastogi ![[assets/slides/vh2VGuQ3zhY/slide-009.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-009.html) - AI slide classifier: `content_slide` confidence `0.99` - Text source: advanced OCR `rapidocr-live/bright-screen/contrast`. - OCR decision: ready — Dense evidence matrix with multiple source callouts and small supporting copy. Slide text: > Teams Are Hitting the Same Tool-Scaling Wall INDUSTRYEVIDENCE > Real evidence from practitioners,engineers,and Anthropic itself,not simulated benchmarks. Ankush Rastogi > tokens 150K → 2K Loading only the toolsneeded for the Anthropic Engineering Blog currenttaskdropsusagefrom1so.oooto 2,000 with98.7%cut. 20+ tools breaksit Vercel AI SDK - Issue #11920 Sending alltool definitions every request choice,and ralses tokers+latency." degrades performance,confuses tool > Official Anthropicfinding Real practitioner issue jon 2026 > Sohnl Shakh > 98% tokenreduction Selects accurately from~3K candidates MCP-Zero (xfey/MCP-Zero) whilecutting tokens 98% Testedon308MCPserversand2.797toob. >2 tools-+stuckloop stuckin aloop,burningthroughallmax iterations.Hit in production. n8n Community Forum "Addmorethan two toolsandtheAlgets > Open source,code cvailable now Production foslure·Aug 2024 > 15 > boseysruy ![[assets/slides/vh2VGuQ3zhY/slide-010.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-010.html) - AI slide classifier: `content_slide` confidence `0.99` - Text source: advanced OCR `rapidocr-live/border-trim/opencv-adaptive`. - OCR decision: ready — Two-column tradeoff table with five rows of small text. Slide text: > KNOWTHE LMITS Trade-Offs & Best Practices > Ankush Rrsi ogi > CONCERN MITIGATION! > Router may miss a needed tool Fallback: raise K, or let the LLM request more: > Adds complexity:vector DB+tuning Embedding search is.ms-fast; the savings dominate: > Rare toois may rank low E Log missesy retrain or add keyword boosting > K threshold is hard to calibrate Start at K=5; tune on a dev eval set. > Not worth it below -20 tools For small tool sets' load staticallypano router > Anbunh Rrtogi ![[assets/slides/vh2VGuQ3zhY/slide-011.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-011.html) - AI slide classifier: `content_slide` confidence `0.97` - Text source: agent_vision. Slide text: > Key Takeaways > 01 Tool Overload Kills Accuracy > 02 Tokens = Money + Latency > 03 Semantic Routing Saves the Day > 04 It's RAG, but for Tools > 05 Start Small, Scale Confidently ![[assets/slides/vh2VGuQ3zhY/slide-012.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-012.html) - AI slide classifier: `content_slide` confidence `0.98` - Text source: advanced OCR `rapidocr-live/bright-screen/contrast`. - OCR decision: ready — Dense reference slide with small multi-column text and many citations. Slide text: > GO DEEPER > Resources & References > Ankush Rstogi > 三 Benchmarks Papers & Tools & Repos API Docs > Zheng et al.-SkillRouter:LLM Agents at Scale(2026) SkillRouter- github.com/alibaba/skillrouter AnthropicMCP- docs.anthropic.com/mcp > Liu&Chen-SemanticTool Selection (vLLM, 2025) FAISS- github.com/facebookresearch/faiss OpenAl Function Calling- platform.openai.com Sohea Shakh > Berkeley Function Calling Leaderboard (BFCL) Pinecone/Qdrant/ChromaDB (vector DBs) Google GenAI Gemini2.0- ai.google.dev > Anthropic Eng.-MCP On-Demand Context(2025) LangGraph-agent orchestration SentenceTransformers-sbert.net > Thank you! > Antush Rastogi ![[assets/slides/vh2VGuQ3zhY/slide-013.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/slides/vh2VGuQ3zhY/slide-013.html) - AI slide classifier: `content_slide` confidence `0.98` - Text source: agent_vision. Slide text: > LET'S CONNECT > Thank You > Scan either code to connect with us on LinkedIn > Ankush Rastogi > Sohail Shaikh ### Hidden Non-Slide Evidence - [`slide-001.jpg`](/assets/slides/vh2VGuQ3zhY/slide-001.jpg) — `other` confidence `0.0`; missing batch classifier result - [`slide-002.jpg`](/assets/slides/vh2VGuQ3zhY/slide-002.jpg) — `other` confidence `0.0`; missing batch classifier result - [`slide-003.jpg`](/assets/slides/vh2VGuQ3zhY/slide-003.jpg) — `other` confidence `0.0`; missing batch classifier result - [`slide-004.jpg`](/assets/slides/vh2VGuQ3zhY/slide-004.jpg) — `other` confidence `0.0`; missing batch classifier result - [`slide-005.jpg`](/assets/slides/vh2VGuQ3zhY/slide-005.jpg) — `other` confidence `0.0`; missing batch classifier result - [`slide-006.jpg`](/assets/slides/vh2VGuQ3zhY/slide-006.jpg) — `other` confidence `0.0`; missing batch classifier result Classification audit: `raw/sources/slide-ai-classification/slides/vh2VGuQ3zhY/audit.json` ## Slide-Derived Subjects To Review Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.