Uploaded July 2026 | Updated September 2026, 2 weeks ago
Company knowledge is scattered across documents, Slack threads, meetings, dashboards, databases, and employees’ heads. When that context becomes stale, AI agents produce bad answers, make poor decisions, and lose user trust.
In this Arize:Observe 2026 session, Rajoshi Ghosh, co-founder and Chief Ecosystem Officer at PromptQL, explains how her team built a company brain that updates itself as people work.
Rather than assigning someone to manually maintain documentation, PromptQL captures decisions, corrections, workflows, and operational knowledge from everyday collaboration. That context becomes part of a living company wiki that AI agents can use across teams and tasks.
Rajo walks through how PromptQL connects to company data, APIs, MCP servers, observability platforms, and coding agents. She also demonstrates how the system helps teams:
• Answer internal questions without repeatedly messaging colleagues
• Track work across teams and notify people when dependencies are resolved
• Capture tribal knowledge and undocumented design decisions
• Debug incidents using code, traces, dashboards, and team context
• Turn conversations and investigations into durable wiki updates
• Detect operational issues and automatically begin incident analysis
• Preserve permissions and access controls across enterprise data
• Build AI systems that improve from shared organizational knowledge
PromptQL’s internal wiki has grown to nearly 3,000 pages, with roughly 200 edits generated each day across a 70-person team.
The central lesson: company context cannot be treated as a one-time documentation project. It has to remain current as a byproduct of how the organization works.
Chapters:
00:00 Why AI agents struggle with stale documentation
00:27 The company knowledge problem
01:48 Tribal knowledge scattered across tools and teams
03:23 How companies try to curate AI context today
04:54 Why context curation cannot be someone’s second job
05:39 Building a company brain that maintains itself
06:52 A 3,000-page wiki with 200 daily edits
07:54 Demo: connecting data, APIs, MCP servers, and coding agents
08:59 Exploring PromptQL’s company knowledge graph
11:02 Replacing internal DMs and “quick questions”
12:23 Tracking work and reducing cognitive overload
14:09 Debugging production systems with shared context
15:18 Root cause analysis across code, traces, and teams
17:18 Turning team decisions into durable knowledge
19:03 Automating incident detection and investigation
20:48 Shared dashboards and team AI workflows
21:56 Why multiplayer context improves AI accuracy
🔗 Learn more about Arize: arize.com
🔗 Explore Arize:Observe: arize.com/observe
🔔 Subscribe for more talks on AI agents, context engineering, observability, evaluation, and production AI:
youtube.com/@arizeai?sub_confirmation=1
#ContextEngineering #AIAgents #EnterpriseAI
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
Company knowledge is scattered across documents, Slack threads, meetings, dashboards, databases, and employees’ heads. When that context becomes stale, AI agents produce bad answers, make poor decisions, and lose user trust.
In this Arize:Observe 2026 session, Rajoshi Ghosh, co-founder and Chief Ecosystem Officer at PromptQL, explains how her team built a company brain that updates itself as people work.
Rather than assigning someone to manually maintain documentation, PromptQL captures decisions, corrections, workflows, and operational knowledge from everyday collaboration. That context becomes part of a living company wiki that AI agents can use across teams and tasks.
Rajo walks through how PromptQL connects to company data, APIs, MCP servers, observability platforms, and coding agents. She also demonstrates how the system helps teams:
• Answer internal questions without repeatedly messaging colleagues
• Track work across teams and notify people when dependencies are resolved
• Capture tribal knowledge and undocumented design decisions
• Debug incidents using code, traces, dashboards, and team context
• Turn conversations and investigations into durable wiki updates
• Detect operational issues and automatically begin incident analysis
• Preserve permissions and access controls across enterprise data
• Build AI systems that improve from shared organizational knowledge
PromptQL’s internal wiki has grown to nearly 3,000 pages, with roughly 200 edits generated each day across a 70-person team.
The central lesson: company context cannot be treated as a one-time documentation project. It has to remain current as a byproduct of how the organization works.
Chapters:
00:00 Why AI agents struggle with stale documentation
00:27 The company knowledge problem
01:48 Tribal knowledge scattered across tools and teams
03:23 How companies try to curate AI context today
04:54 Why context curation cannot be someone’s second job
05:39 Building a company brain that maintains itself
06:52 A 3,000-page wiki with 200 daily edits
07:54 Demo: connecting data, APIs, MCP servers, and coding agents
08:59 Exploring PromptQL’s company knowledge graph
11:02 Replacing internal DMs and “quick questions”
12:23 Tracking work and reducing cognitive overload
14:09 Debugging production systems with shared context
15:18 Root cause analysis across code, traces, and teams
17:18 Turning team decisions into durable knowledge
19:03 Automating incident detection and investigation
20:48 Shared dashboards and team AI workflows
21:56 Why multiplayer context improves AI accuracy
🔗 Learn more about Arize: arize.com
🔗 Explore Arize:Observe: arize.com/observe
🔔 Subscribe for more talks on AI agents, context engineering, observability, evaluation, and production AI:
youtube.com/@arizeai?sub_confirmation=1
#ContextEngineering #AIAgents #EnterpriseAI
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
