Agent Orchestration Platform · Glean Feature Trail

Sep 2026AI Agents for Financial Compliance: How Enterprise AI Is Solving the Regulatory Reporting ChallengeThis post details the application of AI agents to financial compliance, focusing on continuous monitoring, automated audit trail generation, AML/KYC lifecycle management, and policy retrieval. It emphasizes the need for explainability, traceability, and human-in-the-loop controls in regulated environments, outlining a roadmap for mapping compliance knowledge, starting with monitoring, and building an audit story.

Sep 2026From enterprise search to enterprise context: what AI agents actually needThis post details the evolution of Glean's AI agents from relying on enterprise search to requiring a comprehensive enterprise context layer. It outlines the critical requirements for agents: authoritative retrieval, signal over volume, cross-system understanding, permissions-aware context, and context delivery tuned for multi-step work. The post explains how this context layer is built through deep connectors, freshness and indexing, hybrid retrieval, a knowledge graph, permissions enforcement at retrieval, and context delivery for downstream AI. It emphasizes that connectivity alone is insufficient and highlights the measurable benefits of a strong context layer over basic LLM grounding.

Aug 2026Glean Agents can now work independently, build faster, and stay governed at scaleThis post details significant advancements in Glean's agent orchestration platform, focusing on enabling independent agent action, improving agent building workflows, and enhancing governance. Key technical contributions include: the introduction of independent agents with unique identities and provisioned access for workflow ownership and proactive action; support for building agents in external coding tools (Claude Code, Cursor) and syncing via Git; version checkpoints and collaborative editing for agent development; Agent2Agent (A2A) interoperability for inter-platform agent coordination; built-in agent evaluation sets for pre-publishing testing; agent scanning for governance (reviewing instructions, tools, data sources, sharing settings) to flag risks like over-provisioned write tools; and new agent governance policies that combine access, behavior, and usage limits.

Aug 2026Scale AI usage without runaway spend with Glean’s AI usage controlsThis post introduces Glean's AI usage controls, a new capability focused on governing AI spend and scaling usage responsibly. It details the implementation of usage limits that can be set broadly or individually for users, agents, and teams, along with guardrails and access restrictions to models. The post highlights the 'why' behind usage through deep observability into usage details for departments and users, enabling diagnosis of product surfaces and triage of individual user behavior across models, agents, and products. It also introduces the Glean AI Gateway as a mechanism to extend cost governance beyond Glean to other AI surfaces.

Aug 20266 AI agents transforming finance workflowsThis post details how AI agents can be specifically applied to finance workflows, enabling tasks like reconciliation, variance analysis, audit prep, policy adherence, reporting, financial risk assessment, investment research, churn prediction, AP matching, expense planning, and due diligence. It highlights the use of natural language understanding, contextual reasoning, multi-system access, and built-in governance as core capabilities for finance-focused AI agents built on platforms like Glean.

Aug 2026Accelerating the journey to the superintelligent enterprise: New Enterprise Graph, Third-generation Assistant, and Agent superpowers in GleanThis post introduces Glean's third-generation Assistant and Agents, powered by the "Agentic Engine 2". Key technical contributions include: the concept of "vibe coding" for conversational agent building, the integration of over 100 new native actions across various enterprise tools (Slack, Microsoft, Salesforce, Jira, GitHub, Google), and expansion of the action space via the Model Context Protocol (MCP) with new partners. It also details advanced agent capabilities like scheduled triggers, looping, and version control. The post highlights the role of the Enterprise Graph in grounding agents and the introduction of a more powerful Chat API and remote MCP servers for inter-platform agent communication.

Aug 2026Agentic reasoning: The future of Work AIThis post introduces Glean's new agentic reasoning architecture, an evolution from its enterprise search and RAG capabilities. It details the journey from search to Work AI, emphasizing the development of agents that can plan, execute, evaluate, and take actions to achieve complex goals. The architecture involves a decision point between using direct search or agentic reasoning based on query complexity. Agents utilize tools like search, reasoning, and data analysis, and incorporate a 'reflect' step for self-assessment and optimization. The post highlights the potential of agentic reasoning to handle tasks like resolving support tickets and debugging production issues, with early research showing a 24% increase in response and action relevance.

Aug 2026Building an Intelligent Customer Support Agent with Glean APIs and NVIDIA NIM MicroservicesThis post details the integration of Glean APIs with NVIDIA NIM microservices to build an enterprise-grade AI customer support agent. The workflow involves user query contextualization (including image-to-text via NeMo), query guardrailing with NeMo Guardrails, retrieval using Glean's Search APIs (supporting 25,000-character snippets and deduplication), reranking with NeMo Retriever, and answer generation using an LLM NIM (e.g., Llama 3.1) with references from the NeMo Retriever embedding microservice. It highlights deployment flexibility across cloud and on-premises environments and emphasizes Glean's contributions in contextual knowledge graph, permission-aware search, large snippet support, user context, query deduplication, and real-time performance.

Aug 2026Claude Sonnet 4.5 optimizes tool execution to unlock 10% of queries that were previously out of reachThis post details Glean's evaluation of Claude Sonnet 4.5 against Sonnet 4.0, highlighting Sonnet 4.5's ability to tackle an additional 10% of queries previously out of reach. Key improvements are in tool execution efficiency, measured by fewer tool calls (1.61 vs 2.08) and adaptive planning cycles (2.19 vs 3.08). Sonnet 4.5 also shows better parallel tool calling and more efficient parsing of results, committing to answers with higher confidence. This efficiency gain is achieved without degrading correctness or completeness.

Aug 2026Glean Autonomous Agents: Self-evaluating with the context and security you need to automate at scaleThis post introduces Glean's new autonomous agents, which are self-planning, self-evaluating, and capable of solving complex, ambiguous problems. These agents leverage expanded Glean Enterprise Context, which combines memory, connectors, indexes, and graphs to provide a comprehensive foundation for AI. New Glean Protect capabilities for agents are also detailed, including agent moderator roles, access management for actions, agent alignment models, and sensitive data protection. The post highlights improvements in agent completeness (10% higher) and consistency (12% higher) compared to previous generations.

Aug 2026Glean expands horizontal, agent platform with open agent interoperability, dozens of personal and business agents, and comprehensive AI securityThis post introduces Glean's open, horizontal platform for agent deployment and experimentation, featuring an agent builder with natural language input, per-step model selection, and a model hub for integrating various LLM providers. It also details the Agents API, which follows the LangChain Agent Protocol, and provides API client libraries in multiple languages. A hosted MCP server for secure agent interoperability and enhanced agent observability and evaluation tools are also introduced. The release includes dozens of prebuilt quickstart agents for various business functions and personal use cases, leveraging a personal graph. New integrations, such as with Snowflake Cortex Analyst, and companion/sidebar features are highlighted. The agent library aims to scale GenAI tools across organizations. Glean Protect is introduced as an embedded security suite for AI governance, focusing on guardrails, permission rules, and data access controls during agent lifecycle stages.

Aug 2026Glean’s Agentic Engine 2 achieves 94% completeness, excelling at challenging enterprise debugging, writing, and enumeration tasksAgentic Engine 2 introduces adaptive planning and reasoning, moving beyond v1's single planning step to continuously plan and re-plan. It leverages an enhanced Enterprise Graph with 3x more signals, embedded end-to-end. The engine treats third-party agents as callable tools, enabling multi-agent systems. It incorporates both short-term memory for iterative refinement during tasks and long-term memory (Personal Graph and Enterprise Graph) for storing user preferences and organizational patterns. Specific improvements are highlighted in debugging (goal-first planning, artifact-aware code search), writing (improved performance on WritingBench, personalization using Personal Graph), and enumeration/data analysis (scouting for data sources, leveraging Enterprise Graph for hierarchical data, sub-agent delegation).

Aug 2026Glean's MCP servers bring full company context to where your AI runsThis post introduces Glean's Model Context Protocol (MCP) servers, which allow AI and agents to access company-wide context from various tools. It details specific use cases: terminal debugger (parsing stack traces, cross-referencing with code, past errors, and documentation), CI/CD pipeline debugger (integrating Glean search and custom log fetching with IDEs), and documentation writer (using Glean Search for context and publishing to platforms like Confluence). It also highlights the security and control aspects of MCP servers, including centralized management, permission enforcement, and OAuth integration.

Aug 2026How to build an AI assistant for the enterpriseThis post details the challenges and limitations of fine-tuning LLMs for enterprise data, contrasting it with Retrieval Augmented Generation (RAG). It explains why fine-tuning is not suitable for factual recall and introduces RAG as a superior approach. The post elaborates on the critical components of RAG for enterprise AI: a scalable, permission-aware indexing system (contrasting centralized indexing with federated search) and a hybrid search mechanism that combines vector search with traditional keyword retrieval and reranking. It also briefly mentions the role of an extensive knowledge graph for personalization. The technical depth lies in explaining the trade-offs of different AI approaches for enterprise data and outlining the architectural considerations for a robust RAG system.

Aug 2026Introducing contextual image understanding in Glean: The right images at the right timeIntroduces contextual image understanding to Glean Assistant, allowing it to scan connected data sources for relevant images (charts, diagrams, screenshots) and proactively surface them in responses. This involves ingesting images, applying OCR and captioning to build structured metadata, and passing both text and image context to the LLM for improved answer quality and clarity, while respecting user permissions.

Aug 2026Make every business decision a better-informed one with Glean's Deep ResearchThis post introduces the new Deep Research agent, a capability built on Glean's existing agent orchestration platform. The Deep Research agent transforms user queries into systematic investigations by employing research planning, task delegation to specialist sub-agents, parallel fact-finding across internal and external data sources, and synthesizing insights into a structured, citation-rich report. This represents an evolution of the agent orchestration platform by demonstrating its application to complex, multi-step research tasks.

Aug 2026RAG vs. LLM fine-tuning: Which is the best approach?This post provides a comparative analysis of Retrieval-Augmented Generation (RAG) and LLM fine-tuning for enterprise AI use cases. It explains the mechanics, benefits, and ideal scenarios for each approach, emphasizing RAG's strength in providing real-time, contextual relevance and fine-tuning's advantage in domain fluency and task precision. The post also advocates for combining both methods, illustrating with an example of a legal AI assistant. It concludes by positioning Glean's Work Assistant as a platform that supports both RAG and fine-tuned models, enabling tailored AI solutions.

Aug 2026Unlock rich insights across your entire business and BigQuery environments with Glean’s new structured query agentsIntroduces structured query agents that integrate with BigQuery, enabling both direct SQL execution and natural language querying via Google's Conversational Analytics API. This allows users to extract real-time insights from BigQuery without needing deep SQL expertise. The agents can be used for parameterized reports, SLAs, and data-quality checks (SQL) or for ad-hoc questions (natural language).

Aug 2026Using AI Evaluator to ensure Glean Assistant meets modern enterprise needsIntroduced an LLM-based evaluation system (AI Evaluator) to automatically measure the performance of the Glean Assistant. This system evaluates both the retrieval (context relevance rate, context recall rate) and generation (answer relevance, answer completeness, groundedness) components. It addresses challenges like LLM biases and inconsistency by employing techniques such as single-answer grading, categorical rating, prompt engineering, and model ensembles. The system achieved 74% human agreement and improved user retention prediction accuracy by 18.8%.

Aug 2026What is a RAG AI agent?This post introduces and explains the core concepts of Retrieval-Augmented Generation (RAG) and its application in building AI agents. It details the two key functions of RAG: retrieval of relevant, permission-aware information from enterprise sources, and generation by LLMs synthesizing this data. The post outlines the operational flow of RAG AI agents, including query understanding, semantic retrieval, response generation, and optional actions. It highlights the benefits of RAG AI agents for enterprises, such as improved accuracy, reduced hallucinations, enhanced knowledge management, and personalized experiences. Glean's agents are presented as enterprise-grade, secure, scalable, and deeply integrated, offering universal knowledge access, horizontal integration, agentic reasoning, and built-in governance. The post differentiates between single-agent (routing agent) and multi-agent RAG systems, explaining their respective use cases and architectures. It also provides practical guidance on getting started with RAG AI agents, emphasizing starting with a specific use case, connecting the right data sources, choosing a flexible platform, and adopting an iterative approach.

Aug 2026Building agentic reasoning for the modern workspaceThis post details the technical decisions for building agentic reasoning for the enterprise, focusing on the generalization-specialization spectrum of action spaces. It discusses the trade-offs between generalized agents (atomic primitives, high flexibility, higher latency, more errors) and specialized agents (high-level actions, faster execution, consistent outcomes, limited scope). The post also outlines Glean's layered architecture combining search, prompting for business processes, and an agentic reasoning system that can adapt known patterns to novel situations while maintaining alignment with organizational practices.

Aug 2026OpenAI GPT-5 outperforms o3 with 83% correctness, and Glean customers can put it to work todayThis post details Glean's evaluation and integration of OpenAI's GPT-5 model into its agent orchestration platform. It presents comparative metrics showing GPT-5's outperformance over o3 in correctness (83% vs. unknown), completeness, and alignment with human feedback. The analysis highlights GPT-5's broad tool usage and parallel tool calling capabilities as key differentiators, leading to shorter planning steps compared to o3's deep, stepwise approach. The post also discusses GPT-5's improved ability to gauge when to ask follow-up questions and introduces verbosity controls for response length, along with observations on improved formatting quality. The evaluation was conducted using Glean's internal LLM judges and production deployment.

Aug 2026A complete guide to agentic reasoningThis post provides a comprehensive overview of agentic reasoning, defining it as an AI approach that plans, acts, evaluates, and adapts towards a goal. It details the key components (LLMs, tools like web search, code execution, structured memory) and the reasoning loop (task decomposition, delegation, observation, synthesis, adaptation). It introduces agentic reasoning strategies such as Chain-of-Thought, ReAct, Self-reflection, and Multiagent Orchestration, and explains how Glean's Agentic Engine utilizes these for planning, retrieval, generation, and self-reflection, citing a 24% increase in response relevance. The post also discusses real-world applications, benefits, challenges, and getting started with agentic reasoning.

Aug 2026RAG, or Retrieval Augmented Generation: Revolutionizing AI in 2025This post provides a comprehensive overview of Retrieval Augmented Generation (RAG), detailing its fundamental concepts (dense retrieval, sequential conditioning, marginalization), technical architecture (document encoder, query encoder, Transformer-based models, cross-attention), applications (chatbots, content generation, assistive technologies), training and fine-tuning processes (dataset preparation, pretraining, hyperparameter optimization), and challenges (data quality, bias, performance, scalability, ethical considerations). It positions RAG as a breakthrough technology for revolutionizing AI by enabling more informed and contextually relevant outputs.

Aug 2026Agentic RAG explained: Smarter retrieval with AI agentsThis post introduces and explains agentic RAG as a next-generation implementation of retrieval-augmented generation. It details how AI agents are embedded into the retrieval pipeline to enable deeper reasoning, smarter routing, and more adaptive workflows, contrasting it with traditional RAG. Key components like router agents, multi-agent systems, and planning/reasoning agents are described, along with essential implementation aspects such as language model grounding, data infrastructure, agent orchestration, tool integration, and observability. The post highlights agentic RAG's benefits for complexity, scalability, and grounded knowledge, positioning it as a key advancement for enterprise AI.

Jul 2026Agent orchestration platforms comparedThis post compares various agent orchestration platforms, positioning Glean as an enterprise platform that excels in cross-departmental orchestration over company knowledge and business systems. It highlights Glean's strengths in durable execution, branching and looping, routing, scheduled triggers, question routing, approvals, permission-aware context, and trace-level debugging. Glean's openness is noted for its depth of custom actions for specialized engineering workflows and its support for third-party agents and frameworks, including LangChain and LangGraph, via APIs, MCP, and 250+ connectors. The post emphasizes that orchestration is necessary for workflows crossing applications, requiring specialist handoffs, intermediate results, or safe recovery from failures, and outlines key features to look for: workflow control, state management, multi-agent coordination, security, permissions, approvals, observability, auditability, and integrations.

Jul 2026Enterprise context starts with indexing, but it doesn’t end thereThis post expands on the 'Agent Orchestration Platform' by detailing the technical components that constitute Glean's 'system of context' beyond just indexing. It introduces specialized indexing strategies (semantic, lexical, structured) for different data types, the role of the Enterprise Graph in mapping relationships, the function of Memory for persistent learning, the flexibility of Data Connectors for varied data sources, and the necessity of Tools for AI action. It highlights how these elements synergize to provide AI with a deeper understanding of the enterprise, improving accuracy, latency, and token efficiency.

Jul 2026What enterprise voice demands of your tool stackThis post details the engineering challenges and solutions for integrating voice capabilities into Glean's tool stack, highlighting the increased demands on tool selection, execution, and error handling compared to text-based interactions. Key contributions include: emphasizing retrieval quality and precise tool contracts (canonical names, server identity, argument names, schema detail) for voice models with tighter context windows and latency budgets; introducing a filter-rank-trim funnel for dynamic skill and tool discovery; leveraging lexical weighting for disambiguating repetitive tool verbs; implementing structured retry signals for voice validation errors; establishing a canonical execution path for action-oriented work (find skill, resolve tool, validate, execute, confirm); and exposing underlying infrastructure weaknesses (retrieval, initialization, execution-path issues) through voice interactions, leading to improvements in caching, pooling, and offloading setup from the request path.

Jul 2026How to optimize token efficiency in agentic systemsThis post details how to optimize token efficiency in agentic systems by shifting focus from prompt editing to workflow design. It identifies token waste in retrieval noise, bloated working context, unstructured memory, over-orchestration, and unbounded loops. The proposed solutions include measuring token usage by workflow stage, reducing context before it reaches the model via progressive retrieval and selective passing of evidence, replacing raw transcripts with structured state (working state, session summary, long-term facts), routing tasks by complexity, designing loops with clear stopping rules, and reusing proven plans and stable context. Key metrics to track include tokens per successful task, cost per completed workflow, latency per workflow, retrieval precision, groundedness, human correction rate, retry frequency, and context utilization rate.

Jul 2026Agent identity: Agents that act and appear as themselvesThis post introduces and details the 'agent identity' feature for Glean's agent orchestration platform. Previously, agents borrowed the identity of the user invoking them, leading to issues with access control (too broad or too narrow) and audit trails. The new agent identity feature allows agents to act through their own service account credentials, creating a distinct bot/account for each agent across integrated systems like Slack, Jira, and GitHub. This decouples agent invocation from agent permissions, ensures consistent access, and provides proper attribution in audit trails by recording actions under the agent's own account. This also resolves issues with long-running or scheduled agents that were previously hampered by expiring user access tokens.

Jul 2026Building an efficient harness for advanced enterprise workThis post details the re-architecture of Glean's agent harness to use 100% programmatic tool-calling via a Python runtime and tools SDK within an agent sandbox. This shift replaces standard tool calls with code execution, enabling complex orchestration (filtering, looping, branching) within a single script. Key design decisions include using the shell as the only directly exposed tool, exposing other capabilities via the tools SDK, and implementing a sandbox filesystem for intermediate outputs. Tool outputs are truncated in context, with full outputs written to disk. The harness uses search and progressive disclosure to manage tool discovery and context, appending search results to the end of the conversation to preserve prefix cache efficiency. This approach reduced token usage by 24% compared to the previous hybrid model.

Jul 2026Glean AI Gateway: One control plane beneath every AI front doorThis post introduces Glean AI Gateway, a new control plane designed to manage diverse AI front doors and their underlying models. Key technical contributions include: a unified governance layer for model and MCP access, token usage, and security across multiple AI surfaces; cost control mechanisms like budget limits, quotas, and automatic fallback to less expensive models; model agnosticism supporting over 30 providers; an observability layer for tracking token spend, MCP tool usage, model latency, and error rates; consistent application of data permissions and granular access controls; uniform AI security models (prompt injection detection, content filtering) across all front doors and models; and integration with Glean's enterprise graph to provide shared, accurate enterprise context for AI operations.

Jun 2026Buyer’s guide: what to look for in an enterprise AI platform for token-efficient deploymentThis post details Glean's approach to achieving token efficiency in enterprise AI platforms, focusing on platform design rather than solely model choice. It highlights the importance of optimizing retrieval quality (passage-level evidence, relevance, permissions, recency, ownership, metadata, cross-system relationships), context management (selecting only necessary evidence, deduplication, respecting permissions, preferring current sources), and intelligent model routing (using smaller/specialized models for simpler tasks, reserving high-cost reasoning). It also emphasizes testing multi-step orchestration with realistic workflows and verifying observability, governance, and measurement capabilities. Red flags for inefficient platforms are identified, such as a sole focus on models, reliance on oversized context, lack of context narrowing explanation, uniform model path for all tasks, and poor visibility into spend.

Jun 2026Introducing independent agents: AI coworkers securely built for autonomous, multiplayer workThis post introduces 'independent agents,' a new class of AI coworkers built on Glean's context layer. These agents are designed for autonomous, multiplayer work across enterprise applications (Jira, Slack, Teams) while maintaining security and governance. Key characteristics include: identity with context (provisioned access, distinct profiles), memory (learning from company data and interactions), proactivity (acting without explicit prompts, suggesting solutions, sharing learnings), and accountability (traceable actions, tool calls, decisions within Glean's tenant and region, with an emergency stop button). The OnCall Assistant is presented as a specific example, demonstrating how these agents can assist engineers during on-call escalations by reading alerts, gathering context, analyzing logs, and suggesting fixes. The post emphasizes the trade-off between unlocking agent productivity and managing open-ended risk, positioning independent agents as a solution that grounds AI in company context, processes, and permissions.

Jun 2026Your AI coworker in Slack with context across every systemThis post details how Glean's AI coworker in Slack is built on foundational infrastructure that enforces real permissions at index time and during agent actions. It highlights the importance of a unified index for grounding public answers and preventing data leaks. The system's ability to pull context from across multiple systems (Salesforce, Google Drive, Jira, Confluence) to answer complex queries is emphasized. Furthermore, the post discusses the proactive nature of the AI coworker, its ability to chime in and close loops by performing actions like opening pull requests or filing tickets, all under enforced permissions with audit trails. This builds upon the existing agent orchestration platform by focusing on the practical implementation of safe, permission-aware AI actions within an enterprise context.

Jun 2026Beyond prompt engineering: the real drivers of token efficiency in enterprise AIThis post argues that token efficiency in enterprise AI is primarily driven by retrieval quality, context selection, and orchestration design, rather than solely prompt engineering. It details how Glean's Work AI platform leverages Enterprise Context to improve retrieval by connecting knowledge across systems and respecting permissions. The post outlines a four-layer framework for token efficiency: retrieving better and less, passing only relevant evidence through chunking and passage selection, structuring prompts for control, and orchestrating intelligently across workflow steps by summarizing intermediate outputs and avoiding redundant context loading. It also introduces metrics for measuring token efficiency, such as tokens per successful task, latency per task, cost per task, retrieval precision, answer quality, and context utilization rate, and highlights common mistakes like over-focusing on prompt length and assuming large context windows solve underlying system design issues.

Jun 2026AI assistants vs AI agents: What works better?This post clarifies the distinction between AI assistants and AI agents, defining agents as autonomous systems that plan and execute multi-step workflows towards a goal with minimal human input, contrasting them with reactive AI assistants. It details the operational differences, key features like autonomy and tool orchestration for agents, and their suitability for workflow automation. This contributes to the understanding of how Glean's agent orchestration platform enables complex task execution.

Jun 2026Generative AI for software engineers is more than code completionThis post introduces the 'two-layer model' for AI in software engineering, comprising coding surfaces and a shared context layer. It argues that the bottleneck has shifted from code creation to context assembly, leading to an 'AI productivity paradox' characterized by outcome, trust, and safety gaps. The post highlights how a dedicated context layer, like Glean's, connects to core engineering systems (GitHub, Jira, Slack, etc.) to build a permission-aware enterprise graph. This context layer enables AI tools to provide engineers with relevant information for onboarding, daily coding, and incident management, thereby improving efficiency and reducing rework. It emphasizes the importance of trust, governance, and explainability, and proposes a four-question rubric for evaluating AI tools.

Jun 2026Glean MCP Gateway: The context AI needs to get to workIntroduced Glean MCP Gateway, a centralized control plane for managing AI front doors, model access, cost controls, and observability. This gateway enhances security by enforcing permission-aware connectors, IdP-based authorization, granular access controls, and AI security models against prompt injection and malicious code. It enables centralized rollout via MDM and provides detailed usage insights through a dashboard and API. The gateway utilizes precomputed indexes and knowledge graphs for efficient context delivery, improving token efficiency and response quality compared to off-the-shelf MCP tools.

Jun 2026How to build AI chatbots for enterprise teams with GleanThis post details how to build enterprise AI chatbots and agents using Glean's platform. It emphasizes the importance of a strong knowledge foundation, permission awareness, multi-tool context, and action capabilities. Glean provides APIs (Search API, Chat API) and Glean Agents for building these systems. The post outlines a 5-step process: starting with a specific team workflow, getting the knowledge layer right, choosing the build path and action scope, testing before scaling with specific evaluation dimensions (groundedness, instruction following, usefulness, permission correctness, workflow completion), and rolling out by team. It highlights Glean's connectors and its structured understanding of organizational context as key differentiators from generic RAG chatbots.

May 2026Generative AI for software engineers: How to build the right AI stackThis post outlines a layered architecture for generative AI in software engineering, emphasizing the need for a context platform to bridge the gap between coding assistants and broader engineering systems. It details how a context platform, exemplified by Glean, connects code, tickets, incidents, and documentation to provide engineers with grounded, permission-aware answers. The post also discusses the importance of observability AI and agent platforms within this stack, and evaluates AI tools based on their context awareness, trust mechanisms, data governance, and workflow integration. It highlights how this layered approach addresses common adoption challenges like downstream rework, trust issues, and safety concerns by providing a system of context around AI tools.

May 2026Health Agents: Designing a multi-agent platform for proactive health monitoringThis post introduces the Health Agents platform, a system of proactive, dedicated agents built on top of Glean Agents for monitoring infrastructure health across customer deployments. It details the system architecture, which includes raw signals, specialized component health agents, an orchestration and aggregation layer, and user-facing surfaces. Key design tradeoffs include separating component health expertise from deployment-level aggregation and distinguishing lightweight health monitoring agents from deeper debugging agents. The post also discusses mechanisms for building user trust, such as high-quality data, explainability, and feedback loops, and highlights the impact of Health Agents on GTM and Engineering workflows, including customer expansion readiness, faster meeting prep, and cross-deployment health visibility. Future investments include expanding signal coverage, automating agent tuning, and deepening integrations for closed-loop remediation.

May 2026AI tooling stack report for software engineersThis post outlines the emerging AI tooling landscape for software engineers, categorizing tools into IDE coding assistants, enterprise context and knowledge platforms, AIOps/observability assistants, and general-purpose AI platforms. It proposes a two-layer architectural model: an enterprise context layer and coding/interaction surfaces, positioning Glean as the context layer. The post details how AI is being applied in software development workflows (onboarding, coding, incidents) and provides key evaluation questions for AI tooling. It also mentions Glean's integration with IDEs and other tools via protocols like MCP, and its role in providing a unified control plane for AI front doors, model access, cost controls, and observability, aligning with the agent orchestration platform.

May 2026Context makes the Coworker: Glean preferred ~2.5x as often as off-the-shelf MCP tools, which consumed 30% more tokens in Claude CoworkThis post benchmarks Glean's centralized indexing and knowledge graph-based MCP server against federated, off-the-shelf MCP tools using Claude Cowork. It quantifies that Glean was preferred ~2.5x more often and consumed ~30% fewer tokens. The analysis highlights the 'federated search token tax' and the economic trade-offs between upfront data storage costs (indexing) and climbing compute costs (token consumption). It details the evaluation design, metrics (utility, correctness, completeness, tool fidelity), and provides three query examples illustrating the superiority of Glean's context layer in generating work-ready outputs across various task complexities.

May 2026Enable every agent to drive ROI with a robust agent development lifecycleThis post introduces the Agent Development Lifecycle (ADLC) as a structured approach to building and managing enterprise AI agents, mirroring the software development lifecycle. It details the stages of ADLC: Opportunity, Design, Performance, Context, Develop, Launch, and Monitor & Improve. New capabilities are introduced to support ADLC: generally available auto-mode agents for planning and acting without predefined workflows, debug and trace views for visibility into agent execution, sub-agents for modularity, agent sandbox for secure runtime environments with file systems and code interpreters, an agent library for discoverability and governance, agent access policies for organization-wide guardrails, and agent insights for monitoring usage and feedback. These features aim to enable repeatable, measurable outcomes and drive enterprise value from AI agents.

May 2026Best AI tools for software engineers: how to build the right AI stackThis post introduces the concept of an 'AI coding stack' and positions Glean as a foundational context layer within it. It highlights the limitations of current AI coding assistants (like GitHub Copilot, Cursor, Claude Code, Windsurf, and Sourcegraph Cody) in providing organizational context, earning trust, and ensuring safety. Glean's contribution is described as building an enterprise context graph that links code, tickets, incidents, docs, and ownership, enabling permission-aware search and integration with IDEs via MCP. It also mentions Glean agents that can convert tickets and threads into scoped PRs. The post contrasts this with individual coding tools by emphasizing the need for shared, trusted context for organizational coordination.

May 2026Glean Skills know how your business worksThis post introduces Glean Skills, a new capability within Glean's agent orchestration platform. Skills are designed to encapsulate personalized, reusable expertise for enterprise tasks, capturing workflow structure, logic, example outputs, and the tools Glean will use. They leverage Glean's existing enterprise context and can be automatically invoked. Skills also adhere to the open Agent Skills standard, allowing for interoperability with external ecosystems. The post highlights how Skills move enterprise AI from generic assistance to repeatable, personalized expertise, addressing the challenge of operationalizing AI for specific business processes like account planning or executive business review preparation.

Apr 2026The harness as the context managerThis post details four key changes in Glean's agent harness design to improve long-running agent performance and context management: 1. Programmatic Tool Calling (PTC) in sandboxes moves workflow logic into sandboxed code, keeping intermediate state in variables and files, enabling batching, parallelization, and more robust error recovery. 2. Sub-agents decompose agent loops into isolated execution contexts, each with its own context window, allowing for parallel analysis of large collections of items. 3. Compaction strategies preserve load-bearing conversation state while moving raw intermediate outputs to file systems, addressing context window limits over time. 4. Search-first discovery of skills decouples tool and skill discovery from schema hydration, fetching full definitions only at execution time by indexing skills and ranking them using enterprise graph signals. These changes collectively enhance context management, scalability, reliability, and efficiency for agents performing complex, multi-step tasks.

Apr 2026Glean Waldo: An agentic search model that pairs with LLMs to deliver frontier intelligence with ~50% lower latency and ~25% fewer tokensIntroduced Glean Waldo, a specialized agentic search model built on NVIDIA Nemotron 3 Nano, designed to optimize enterprise AI search latency and token usage. Waldo acts as an initial planning agent, executing tool plans (Glean Search, employee search, web search) before invoking a frontier model. This architecture reduces end-to-end latency by ~50% and token usage by ~25% compared to using a single frontier model for both planning and reasoning. Waldo is trained using Direct Preference Optimization (DPO) on production tool-use patterns and reinforcement learning with a task-specific reward signal based on retrieved document relevance. The post also details how Waldo's execution signals are used to enable adaptive reasoning, dynamically determining the required reasoning depth for a query, further contributing to efficiency.

Apr 2026Why the AI stack for modern engineering teams requires both coding and contextThis post introduces a two-layer model for AI stacks in engineering, emphasizing the separation of coding assistants (IDE-based) from enterprise context platforms. It argues that while coding assistants are crucial for raw output, the real productivity bottleneck is chasing context across various tools. The proposed solution involves a foundational context layer that indexes and provides information across teams and repos, including incidents, tickets, and design docs, while also enforcing security and privacy policies. This layer enables AI tools to provide context-rich, compliant results. The post highlights three key trends: AI moving from experiments to infrastructure, the need for blended tools rather than single vendors, and the increasing importance of security, governance, and explainability.

Mar 20267 Core Components of an AI Agent Architecture ExplainedThis post details the seven core components of an AI agent architecture: goal definition, perception and input processing, memory systems, reasoning and planning, tool execution and action, orchestration and coordination, and feedback and observability. It emphasizes enterprise governance, security, and auditability, highlighting how each component contributes to building scalable and trustworthy AI systems. Specific technical considerations include input sanitization, multi-modal awareness, hybrid memory models, various reasoning strategies (rule-based, chain-of-thought, tree-of-thought, Monte Carlo), action sandboxing, and observability mechanisms for agent performance and cost. The post also references Glean's existing capabilities in unifying perception pipelines, offering context-aware access to enterprise knowledge, and ensuring consistent control when agents act across systems.

Mar 2026Trace learning for self-improving agentsThis post introduces Glean's approach to trace learning for self-improving AI agents. It details a two-component system: offline learning to mine strategies from historical traces and online application to retrieve and apply those strategies at runtime. The offline process involves a 'teacher' agent and multiple 'student' agents to score and distill effective strategies. The online application retrieves relevant memories to adapt agent plans before execution. The post emphasizes designing for enterprise accuracy by extracting assertions across multiple responses, verifying conflicts with Glean Search, and avoiding learning when inconsistencies arise. It also highlights the importance of storing intentionally narrow tool strategies that generalize across related tasks and splitting learnings into deployment and user levels for security and personalization.

Mar 2026Is MCP dead? When to use MCP for enterprise AIThis post clarifies the enterprise value proposition of the Model Context Protocol (MCP) by distinguishing between local, single-user workflows and company-scale AI deployments. It argues that for enterprises, MCP serves as a crucial platform layer for centralized auth and secrets management, org-wide telemetry and observability, standardized prompts/skills and docs across surfaces, and a single integration surface for multiple hosts. It highlights that MCP enables AI to be run as a governed, observable service for thousands of people, contrasting this with simpler CLI or ad-hoc wrapper solutions suitable for individual developers. The post emphasizes that MCP's value lies in its ability to provide a common contract between AI hosts and tools, manage auth and policy centrally, and deliver skills and docs consistently across various AI surfaces.

Mar 2026LLM choice in Assistant: A better agentic coworker sourcing deep proficiencies from your models of choiceThis post introduces LLM choice within Glean Assistant, allowing users to select AI models optimized for specific tasks (e.g., coding, creative assets, general purpose). This feature enhances flexibility by enabling teams to adapt to new models and their varying strengths, cost profiles, and tradeoffs without switching tools or reconfiguring systems. For administrators, it provides granular control over model usage, policy definition, and access governance, ensuring enterprise oversight and compliance while users benefit from task-specific AI proficiencies.

Mar 2026Context engineering AI: The foundation of reliable, high-performing modelsThis post introduces and defines 'context engineering' as a critical discipline for building reliable, high-performing AI applications and agents. It details the four core layers of context (content, structural, task, activity) and the specific inputs to the LLM's context window (system instructions, user input, short-term memory, long-term memory, RAG, tool calls, structured output formats). It contrasts context engineering with prompt engineering, highlighting its importance for dynamic, repeatable workflows. The post also outlines the benefits of strong context engineering (higher accuracy, faster outputs, aligned decision-making, lower costs, safer AI) and the risks of poor context engineering (context poisoning, distraction, confusion, and poisoning).

Mar 2026Introducing MCP in Glean: From connectivity to enterprise-ready actionsThis post introduces the integration of the Model Context Protocol (MCP) into Glean's agent orchestration platform. It details how MCP enables AI agents to connect to and take actions in enterprise systems. Technical contributions include the development of tool search to narrow down action candidates for models, the use of enterprise memory to capture and surface successful agent execution traces, and the definition of reusable skills as executable markdown specifications for complex workflows. Security enhancements focus on centralized governance of MCP servers, enforcing least-privilege execution for every action, and scanning MCP inputs/outputs for adversarial patterns.

Feb 20268 AI agents to help software engineering teams ship and fix fasterThis post introduces eight new AI agents designed to accelerate software engineering workflows. These agents target specific pain points in the engineering lifecycle, including: 1. Pull request review: scans code changes and suggests improvements. 2. Resolve GitHub PR feedback: aggregates and prioritizes unresolved comments. 3. Resolve Jira ticket: analyzes ticket context and suggests next steps. 4. Spec to implementation PR: guides implementation from design docs. 5. Engineering project onboarding: surfaces key project information and contacts. 6. Launch documentation: assembles release notes and supporting docs. 7. Engineering self-evaluation: generates structured evaluations from contributions. 8. Engineering standup: generates daily updates from work artifacts. These agents are integrated into Glean's existing agent library and are part of the February Drop.

Feb 2026Over 85 new actions enable you to get more done with better AgentsThis post introduces over 85 new actions for Glean's agents, significantly expanding their capabilities across various enterprise applications (Salesforce, Jira, GitHub, Confluence, Google Calendar). These actions enable agents to perform complex tasks like creating/updating opportunities, managing tickets, reviewing pull requests, creating documentation pages, and scheduling meetings. The post also reiterates the concept of agent sandboxes, which provide agents with a virtual environment to handle larger datasets and complex analyses, and emphasizes that all actions operate within existing company permissions and security models. A practical example illustrates an agent investigating an escalated Jira ticket, analyzing codebase context, updating the ticket with structured information, and documenting the solution in Confluence.

Feb 2026Why agents need sandboxes, not larger context windowsThis post introduces the concept of agent sandboxes as a solution to the limitations of LLM context windows for long-running agent tasks. It details how sandboxes provide agents with a virtual computer environment (file system, command line, code runtime, tool index) to expand their effective short-term memory. The post explains how sandboxes enable agents to process larger datasets, perform complex analyses (e.g., pipeline health analysis, sensitivity analysis), and overcome context overload by storing data on the filesystem for on-demand access. It also discusses the engineering lessons learned from previous memory management strategies and highlights the security aspects of sandboxes, including isolation, resource limits, and permission inheritance. The core technical contribution is the architectural shift from context window truncation to persistent storage in a sandboxed environment for agent memory management.

Feb 2026Not all enterprise context is created equal: Glean results are preferred ~2× more than ChatGPT and 1.6× more than ClaudeThis post details an evaluation comparing Glean's enterprise context layer against OpenAI ChatGPT's company knowledge and Anthropic Claude's enterprise search. It found that human graders preferred Glean's answers 1.9x more often than ChatGPT's and 1.6x more often than Claude's on correctness. The post explains the challenges of context rot in LLMs and the importance of a robust enterprise context layer for agentic work. It outlines the evaluation design, including the queryset, dataset construction (using Glean's internal deployment as a proxy), and grading methodology (blind, randomized evaluation on correctness and completeness). The results show Glean's strengths in finding canonical documents, understanding jargon, handling ambiguity, and answering technical queries, attributed to its semantic search models and Enterprise Graph. It also contrasts Claude's 'deep' sequential tool-calling strategy with ChatGPT's 'wide' parallel search strategy, highlighting potential inefficiencies and inaccuracies in both. The post emphasizes the persistence of indexes and knowledge graphs due to their accuracy and avoidance of brute-force compute.

Feb 2026Glean skills: A new way to package and reuse enterprise expertiseThis post introduces Glean's support for the Agent Skills standard, enabling users to package and reuse enterprise expertise. It details how skills are structured (description, instructions, resources), the importance of negative examples and few-shot examples for skill routing and accuracy, and the benefits of skills for multi-tool orchestration. The post also discusses the symbiotic relationship between skills and agents, with skills acting as units of expertise and agents as orchestrators. Glean plans to automatically build skills by analyzing user behavior, making expertise a first-class citizen in the AI stack.

Feb 2026The emerging agent architectureThis post outlines Glean's strategic approach to building an open and horizontal agent architecture. It details the evolution of the agent stack, emphasizing the importance of context engineering, model agnosticism, robust orchestration, unified security, and diverse interfaces. Key technical concepts discussed include the role of context graphs, the separation of context from models to prevent vendor lock-in, the economics of frontier model development, and the feedback loop between context and orchestration for reliable automation. The post highlights the shift from individual agent experiments to operating hundreds of agents within an organization, requiring a scalable and interoperable foundation.

Feb 2026How do you build a context graph?This post details Glean's approach to building context graphs, which are essential for AI to understand and automate work processes. It defines context graphs as models connecting enterprise entities with temporal traces of actions and events. The post outlines the 'how' of building these graphs, emphasizing deep connectors, observability, a unified knowledge graph, personal graphs, and the creation of anonymized, probabilistic process traces. It highlights the hybrid storage model using embedded entity IDs for step-by-step traversal and the importance of closing the loop with agentic traces to continuously learn and refine the context graph. The post also mentions an internal test of the concept using opt-in personal graph data.

Jan 2026Glean Agents: the horizontal environment for building, deploying, orchestrating, and governing AI agents in the enterpriseThis post announces Glean Agents, a horizontal environment for building, deploying, orchestrating, and governing AI agents in the enterprise. It details the capabilities of the agentic reasoning engine, which executes multi-step workflows by retrieving data from various sources and using actions. The post highlights the agent builder, which allows users to express agent functionality in natural language, supporting intelligent branching and agentic looping. It also introduces an agent library for discovering and sharing pre-built agents and emphasizes the platform's LLM-agnostic nature and commitment to an ecosystem of agents through APIs and partner integrations. The post also mentions the ongoing evolution of agent orchestration to support third-party agents and frontier GenAI developments.

Jan 2026Glean introduces Glean Agents and advances Work AI platform with the most comprehensive access to work and world knowledgeThis post introduces Glean Agents, a new horizontal environment for building, deploying, orchestrating, and governing AI agents within Glean's Work AI platform. It highlights the ability for users to create agents using natural language descriptions, powered by Glean's LLM-agnostic reasoning engine. Key new capabilities include universal knowledge access (integrating enterprise data with web and LLM data), structured data analysis (querying live data from applications like Salesforce, Databricks, and Jira), and active data and AI governance (proactively scanning and remediating overshared sensitive data). The post emphasizes the scalability of agent deployment and management, aiming for enterprises to manage thousands of agents securely and compliantly.

Jan 2026Glean's open agent platform: because while not all agents will be built on Glean, all agents should be connected to enterprise dataThis post introduces Glean's support for universal communication protocols like Anthropic's Model Context Protocol (MCP), LangChain, and LangGraph. It details how Glean can act as a host to connect to external tools via MCP servers, enabling Glean agents to call external services. It also describes how external tools, such as Cursor, can call Glean agents to access enterprise context. The post emphasizes the value of a common communication framework for agent interoperability and avoiding vendor lock-in.

Jan 2026Context is the next data platform—and why context graphs are key to understanding processesThis post introduces the concept of context graphs as a key component for enabling AI agents to understand and automate enterprise processes. It details Glean's approach to building context graphs by focusing on capturing the 'how' of work through detailed observability of activity data across various enterprise tools. The post explains the technical challenges involved, such as deriving higher-level understanding from atomic actions and handling privacy constraints. It positions context graphs as part of a foundational technology stack for agentic automation, working alongside connectors, indexes, and memory to create a new data platform.

Dec 2025Shifting the work of context engineering to the AI platformThis post details Glean's design decisions and lessons learned in building enterprise context for AI agents. It introduces the concept of shifting context engineering from individual engineers to the AI platform. Key technical contributions include: 1. **Index Diversification:** Separating indexes for different jobs (e.g., calendar search by time/people, Glean search for unstructured data, Salesforce dashboard index for metrics) to improve retrieval. 2. **Hybrid Connectors:** Combining MCP tools with Glean's indexing system for real-time freshness and unified understanding, emphasizing the need for ongoing data modeling and signal extraction for MCP tools. 3. **Knowledge Graphs:** Using graph-based structures to encode entities (people, projects, processes) and their relationships for multi-hop reasoning, disambiguation, and understanding work flows. 4. **Tool Search:** Creating an index of tools and learning from successful agent workflows to map common patterns and select the best next tool, optimizing for speed and reducing hallucinations. 5. **Enterprise Memory:** Storing traces of agent runs (tools used, parameters, sequences, outcomes) and offline permutation analysis to learn tool-use heuristics, optimize tool parameterization, and improve action sequencing. This memory is fed back into the system to guide tool selection and structuring.

Nov 20252025 search tool benchmark: key metrics to evaluate accuracy and speedThis post details key metrics for evaluating enterprise search tools in 2025, focusing on accuracy, speed, user experience, and pricing. It introduces specific benchmarks for tool calling accuracy (90%+), context retention (90%+), and response times (<1.5-2.5 seconds). It also discusses update frequency, interface usability, reporting quality, and compares Glean with other leading search tools based on these metrics. The post emphasizes aligning search tool capabilities with business needs and use cases, particularly for AI-powered search platforms.

Nov 20255 metrics to measure AI-generated answers' decision-making impactThis post introduces a framework for measuring the decision-making impact of AI-generated answers, focusing on five key metrics: accuracy, relevance, coherence, helpfulness, and user trust. It details how these metrics can be systematically measured and tracked to quantify improvements in decision speed, quality, and confidence, transforming AI from a promising technology into a measurable driver of better business outcomes. The platform's retrieval augmented generation (RAG) architecture grounds answers in verified company data, ensuring compliance and security.

Nov 2025The all-new Glean desktop app: Frictionless, instant answers anywhere you workIntroduces a new surface for Glean's AI Assistant: a native desktop application. This app provides full context across all open windows, unlike browser extensions or mobile apps. Key new features include: 1. Instant keyboard shortcuts (Command-Shift-J) to summon the Assistant without leaving the current application. 2. An 'always-on-top' quick chat window for side-by-side interaction and iteration. 3. Screenshot-to-chat functionality, allowing users to share visual context (documents, UI states) for more relevant answers. The post also reiterates Glean's commitment to privacy and control, stating the app respects enterprise permissions and does not index local files.

Nov 2025Glean Agents adapt to real-world complexity and are built to scale safely across your enterpriseThis post introduces a unified agent builder that merges graph-based and conversational approaches, allowing for natural language creation and step-by-step manual control. It also adds 'Fast' and 'Thinking' modes to the plan & execute step, enabling agents to optimize for speed or depth of reasoning, with 'Thinking' mode achieving 94% end-to-end answer completeness in internal testing for complex use cases. Granular agent permissions are introduced for admins to control sharing and publishing, and agent routing from Glean Assistant is enabled via configurable rules to direct user queries to specialized agents.

What this trail covers