Mapping workflow_
Mapping workflow_
Blog / Insights
We distill global AI developments and translate them into operational and business context for Indonesia.
Move beyond policy documents. Build a production AI agent governance framework that connects authorization, audit evidence, monitoring, ownership, release gates, and cost controls.
Turn AI agent traces into production controls with practical SLOs, alert routing, incident response, privacy safeguards, and safe rollback patterns.
Design an AI agent approval workflow with risk-based gates, useful reviewer context, durable pauses, escalation, audit trails, and measurable controls.
Memory is durable context that an agent can retrieve across runs to guide its behavior. It may include facts, preferences, past interactions, instructions, skills, examples, and learned patterns. A...
Most AI pilots don't die because the model was wrong. They die because no one built a path from "it worked in the demo" to "it runs reliably on Monday morning." A production roadmap for enterprise AI is not a technology strategy document—it
At its core, an agent is just a model calling tools in a loop until a task is complete. .png) This is what LangChain’s createagent gives you. Pick any model, plug in tools, and you have a working...
Imagine you are back from a week of vacation and need to catch up on Slack threads, GitHub PRs, Linear tickets, and calendar invites. The work spans several tools, requires judgment, and may produce...
- LangSmith processes billions of tokens a day across production traces. One of our core challenges is efficiently mining signals across these traces - We partnered with Fireworks to build an...
SmithDB supports full-text search and JSON filtering over agent traces with a median (P50) latency of 400 ms, even though the underlying data consists of large, deeply nested JSON documents stored in...
- Most agent tools only see the backend. Browsers, apps, and devices contain valuable state and capabilities that traditional server-side tools cannot directly access. - Headless tools bring...
Think about what Cursor, Claude Code, or ChatGPT's code interpreter can do that a plain chat interface can't. They don't just answer questions: they run the code, see the error, fix it, run it again,...
Transient failures are the most common kind of failure in any non-trivial graph: an LLM provider returns a 5xx, a vector store hits a connection reset, a downstream HTTP service is briefly...
Every twenty years or so, the way software gets built changes in a way that forces every engineering organization to restructure how they work. On-prem to cloud was the last one. Agents are the next...
- A harness is the scaffolding around the model that connects it to the real world. - How well a harness fits the task at hand determines how useful an agent is. - LangChain's createagent is the...
- Verifiers can be a cost bottleneck for running agent evaluations and RL post-training at scale. - We find we can reduce verifier costs by an order of magnitude by batching verifiers and using open...
- Lyft moved agent development closer to the people who understand customer issues best. By letting ops teams, VoC leads, and product managers define agents through prompts and configuration, Lyft...
.png) Improve agents without spending hours buried in traces. LangSmith Engine watches your production traces, clusters recurring failures into named issues, diagnoses root causes, and proposes fixes...
- Mission Control reduces the operational sprawl around self-hosted LangSmith. Platform teams can manage configuration, preflight checks, health, release history, diagnostics, and support workflows...
- Streaming needs to evolve beyond tokens Modern agents generate messages, tool calls, subagent activity, state changes, approvals, and media, requiring structured event streams instead of flat text...
- Agent sandboxes need stronger network controls because agents can run code, install packages, call APIs, and follow instructions from untrusted content. - LangSmith Auth Proxy keeps credentials...
- Interpreters sit between serial tool calls and full sandboxes. Agents get code-level composition over scoped capabilities without inheriting a whole environment. - Interpreter state is a third...
LangSmith is the home of the agent improvement loop. Build, test, deploy, and monitor are the four pillars of this loop that power agent development. As the number of agents you deploy grows, the...
- LangChain Labs is focused on continual learning for agents. The goal is to help agents improve from the data they already produce, including traces, feedback, eval results, and production behavior....
- Accelerating the agent development lifecycle. LangSmith Engine watches your production traces, clusters failures, and opens PRs with fixes so you can spend time reviewing improvements, not hunting...
- Agent traces have outgrown traditional observability stores — modern agent traces contain hundreds of nested spans, multi-modal content, and spans that stay open for hours, creating data volumes...
- The agent development lifecycle now moves faster. LangSmith Engine replaces the manual cycle of reading traces, spotting patterns, and writing fixes by doing it continuously — clustering production...
- Building agents is getting easier. Operating them is still the hard part. Long-running agents need durable execution, tool access, sandboxes, memory, and tracing, and assembling that yourself takes...
- LangSmith Sandboxes are now GA — Each sandbox runs as a hardware-virtualized microVM, fully kernel-isolated from your services and other sandboxes, making them genuinely secure for running...
- Context shapes agent behavior. Alongside the model and code, context - instructions, skills, and policies - has an outsized impact on how agents perform. Many failures trace back to context that's...
