LLM+P turns LLMs into reliable natural-language front ends for proven symbolic planners. That reduces execution risk and often lowers real-world costs (e.g., fewer extra robot trips). It avoids expensive LLM fine-tuning by delegating correctness to existing planners.

If you build assistants that call external services, training on many real APIs plus a retriever and multi-path planning dramatically reduces manual engineering and makes open-source models practically competitive with closed systems.

If your product must use live or private data, you need tested tool integration and source selection; relying on a base LLM risks wrong or outdated answers.

OptiGuide speeds what‑if and root‑cause analysis for planners, reduces engineer on‑call cycles, and keeps sensitive data in‑house while surfacing solver decisions in plain English.

API-Bank lets product and engineering teams measure how reliably models call real APIs, cheaply create high-quality tool-use training data, and reduce costly manual labeling. Improving API reliability cuts user-facing errors like failed requests or wrong side effects.

Agentic LLMs that run verified code and fetch trusted web facts can unlock personalized insights from wearable data—improving product value for health apps while reducing numeric errors and buggy analyses.

SheetCopilot lets non-technical users automate many spreadsheet tasks by speaking plain English, lowering manual work and reducing mistakes, but it still needs human verification for critical data because full correctness is about 44% on evaluated tasks.

Train web agents from their own search traces to get large, fast gains in task success without risky online RL; pairing a trained policy with online search gives near‑perfect results for structured web tasks.

CoA makes multi-step tool use both more accurate and faster by separating plan generation from tool calls; this reduces arithmetic bugs and shortens latency when pipelines must call external APIs.

SWIFT unifies fine-tuning, RL-style alignment, quantization and deployment for text and multimodal models. That reduces engineering overhead, accelerates experiments on agents and lets teams run many models and tuners without building custom glue code.

AutoFlow reduces manual workflow design time by automatically producing readable, executable agent workflows and can raise task performance on image/text benchmarks, lowering operational cost for multi‑step agent tasks.

Linking LLMs to real tools turns language models from static answerers into actionable assistants that fetch fresh facts, run domain tools, automate workflows, and show decision steps—improving trust and utility in products.

Automates converting published code into reusable tools, cutting expert setup time and enabling rapid integration of domain-specific methods into production agents at a low per-tool cost.

T-Eval gives runnable, per-skill diagnostics for building LLM-based tool agents so teams can pinpoint whether problems come from planning, choosing tools, formatting requests, or checking results.

If your product needs reliable multi-step interactions with many third-party APIs (search, image tools, web services), a small router + summarizing state manager can boost success and reduce repeated failures with little engineering overhead.

EHRAgent reduces dependence on data engineers by letting clinicians ask EHR questions in plain language and getting accurate answers; this can speed workflows but increases runtime/API calls and needs privacy safeguards.

Tool-using LLM agents can make costly or unsafe actions; organizations should not deploy them without runtime checks, human-in-loop gating, and improved evaluation tailored to agent behavior.

If you plan to automate user tasks with LLMs, expect frequent multi-step failures and risky incorrect side effects; instrument tool calls and add verification before irreversible actions.

If you want LLMs to make real product recommendations in a specific domain, wrap them with a KB retriever and a goal planner; that combination turns an LLM from brittle zero-shot text generator into a materially better recommender on evaluated datasets.

SoAy converts complex multi-API academic queries into a short plan plus executable code, cutting latency and improving answer reliability—useful for search services, institutional dashboards, and any product that needs precise, API-backed facts.

ToolSandbox tests realistic, multi-turn tool use and highlights where agents hallucinate, fail to sequence dependent actions, or mis-handle time—insights you need before putting LLM agents in customer-facing automation.

If you must run KBQA with limited labeled data, KBQA-o1 lets you use open LLMs and automated MCTS exploration to generate high-quality training pairs and sharply improve accuracy on complex queries.

You can convert repeated human guidance into model updates that reduce prompt length, cut inference cost, and raise multi-tool task reliability with modest annotation work.

Agentic harmonization can speed up combining heterogeneous datasets and produce reusable, publishable transformation scripts that improve reproducibility and reduce manual engineering time.