Editorial illustration for Industry Shifts to Richer Context for AI Agents, Guided by Human Judgment
AI Agents Get Smarter with Rich Context and Human Guidance
Industry Shifts to Richer Context for AI Agents, Guided by Human Judgment
Why does the way we feed AI matter? In the first wave of autonomous assistants, developers handed models a lone system prompt and a handful of tool definitions, hoping the algorithm could fill in the gaps. That minimalist approach left agents wandering through tasks with little sense of the surrounding workflow.
As startups and larger firms began to stitch AI into real‑world processes, the shortcomings became obvious: without a broader frame of reference, agents missed cues, duplicated effort, and often required costly human overrides. The industry’s response has been to embed richer, pre‑execution context—essentially giving the model a concise briefing before it starts work. This shift aligns with a growing emphasis on human judgment within the improvement loop, where engineers curate the information that steers the agent’s decisions.
One emerging standard, Anthropic’s Skills, has quickly gained traction since its launch, signaling that richer context is no longer optional but expected.
*Agent Context*—...
Agent Context Early agents just gave the model a single system prompt and a set of tool definitions. Over time, the industry has moved toward providing agents with much richer context at the beginning of their execution. Anthropic's Skills, a standard that has quickly grown in popularity since launching in October, is one prominent example of this trend.
Instead of cramming everything into one system prompt, your team curates documentation, examples, and domain rules in advance, then lets the agent fetch what it needs at runtime. This lets the agent use far more knowledge without bloating the system prompt. Effective agent design involves deciding what knowledge the agent should access and organizing it so the agent can retrieve the right information at the right moment.
At minimum, our trader copilot needs to know how to use the database and understand its schema. Depending on the nature and amount of additional knowledge from our team that our copilot needs, we'll have to spend time not just collecting that knowledge but determining how to structure and progressively disclose it to our agent. Choosing and structuring the information available to the agent when it starts up is part of the discipline of context engineering.
Context engineering also covers how the information you provide in each LLM call evolves as the agent moves through its task. The feedback your human stakeholders provide when reviewing your agent's outputs and evaluation scores may influence how you approach end-to-end context engineering for your agent. Now that we have outlined the parts of an agent that benefit from human judgment, we'll cover how to collect that human input.
Incorporating human judgment into the agent improvement loop At LangChain, we've worked with hundreds of organizations deploying AI agents.
Human judgment still drives AI agents. Early versions relied on a single system prompt and a handful of tool definitions; today, providers feed agents richer context from the outset. Anthropic's Skills standard, quickly adopted since its launch, exemplifies this shift.
Yet organizations must translate tacit knowledge—insights that live inside employees—into a form agents can consume. The article notes that many teams only recognize the value of that hidden expertise when they attempt automation. Without clear methods for extracting and codifying such wisdom, the promised gains may remain limited.
Moreover, it is unclear how consistently richer context improves task outcomes across different domains. The move toward richer initial context reflects a broader belief that context matters, but evidence of measurable impact is still emerging. Will this added depth translate into consistent performance gains?
As firms experiment, they will need to monitor whether the added information translates into more reliable or efficient agent behavior, or whether the overhead of gathering and maintaining that context outweighs benefits. The balance between documented knowledge and elusive tacit insight will likely shape future agent designs.
Further Reading
- Agents Accelerate the Next Wave of AI Value Creation - BCG
- Part 1: Agentic AI - A Paradigm Shift for Integrated Business Planning - Board
- The essential guide to scaling agentic AI - IBM
- AI Agents Changing Every Industry: Reshaping the Global Economy - Klover
- Agent-Supported Foresight for AI Systemic Risks - arXiv
Related Reading
Business & Startups OpenAI, a Series F San Francisco startup founded in 2015 by eight pioneers
Business & Startups Terminal-Bench 2.0 launches with Harbor, testing any container-installable agent
Business & Startups Zuckerberg Unveils Meta Compute to Build Global AI Infrastructure
From Archives Google unifies Gemini Enterprise Platform and Application in new release
From Archives Mars leverages Gemini Enterprise to build AI agents accessing century‑old data
Common Questions Answered
How have AI agents evolved in their approach to context and task execution?
Initially, AI agents were given minimal context through a single system prompt and limited tool definitions, which led to inefficient task navigation. The industry has since shifted towards providing agents with richer, more comprehensive context upfront, including curated documentation, examples, and domain-specific rules.
What is Anthropic's Skills standard and why is it significant?
Anthropic's Skills standard, launched in October, represents a new approach to AI agent context by moving away from cramming everything into a single system prompt. Instead, it allows teams to proactively curate and provide detailed documentation, examples, and domain rules, which has quickly gained popularity in the industry.
Why is human judgment still crucial in developing AI agents?
Human judgment remains essential because AI agents need help translating tacit knowledge—insights that typically reside within employees' expertise—into a format that machines can understand and utilize. Organizations often only recognize the value of this hidden expertise when they attempt to automate processes, highlighting the ongoing need for human oversight and input.