For Balluff

Agentic AI in Project Management

More room for the essence of good project work.

Project lifecycle

Leadership shifts across phases

Phase 1 · Idea

PM leads, AI asks for context.

Phase 2 · Concept

PM leads, AI finds gaps.

Phase 3 · Planning

PM leads, AI checks dependencies.

Phase 4 · Execution

AI drives operations, PM decides.

Phase 5 · Benefits

PM leads review, AI distills learning.

Change requests, new insights, and external impacts flow back into context.

Getting started

More room for the essence

  • Clear project status from meetings, board, and documentation
  • Faster decisions through better questions
  • Less search time, more focus on value creation

Discussion: Where does friction arise today?

Daily work

Where time is lost

Catching up on status

Aligning status between stakeholders — after every meeting and email.

Consolidating sources

Merging information from meeting, email, and board into one picture.

Checking reality

Reconciling board status with what actually happens in day-to-day project work.

Question: What happens today between meeting and board?

Lever

Meeting transcripts as raw material

  • Status meetings
  • Alignment sessions
  • Technical meetings
  • Every level can contribute project context

Shared AI project context in the Microsoft environment — a pragmatic starting point for Balluff.

Context

From transcript to project memory

Structure

  • Summary
  • Open items
  • Risks

Decisions

  • Decisions made
  • Change requests
  • Open questions

Board link

  • Epics
  • Features
  • Tasks
Guardrail

Azure Boards read-only first

  • PAT read-only only
  • API access local and cache-aware
  • Analysis as the first step
  • Writing later with proxy, MCP, sentinel, and confirmation

Note: Write access can damage boards quickly — read-only protects the shared truth.

Analysis

What agents can detect

  • Gaps between rough and detailed concept
  • Discrepancies between meeting and board
  • Missing acceptance criteria
  • Open delivery commitments
  • Scope drift and conflicting status signals
  • Missing decision basis for the next meeting
Assistance

Agent as question engine

  • What is missing for the next decision?
  • Which assumption is untested?
  • Where does board status diverge from the discussion?
  • Which risks need re-evaluation?
  • Which question raises the maturity of project context?
Project knowledge

RAG and LLM-Wiki

  • Project management handbook
  • Meeting history
  • Roles and contacts
  • Technical documentation
  • Decisions and open questions

Vector-RAG

Semantic search across documents and transcripts.

Graph-RAG

Relationships between epics, dependencies, and roles.

Hybrid

Semantics and structure together — for complex project questions.

LLM-Wiki (Karpathy): growing project memory — each session enriches the context.

Automation

Generate decks from context

  • Presentations generated automatically from the accumulated project context
  • Context is enriched in a value-creating way — not a one-off artefact
  • No knowledge left stranded in a deck that sits unused afterwards
  • Status report, meeting brief, stakeholder update: all from one source

This deck is an example: it can be regenerated from the shared project context at any time.

Next step

Start small and learn fast

  • Start now with a small pilot
  • Transcribe meetings and maintain context deliberately
  • Analyze Azure Boards read-only
  • Generate a question list for the next meeting
  • Iterate quickly when productivity gains are tangible

As context grows, the abstraction level of conversations rises.

Appendix

Case study: mechanical engineering project

  • Meeting transcripts related to Jira user stories
  • Gaps and discrepancies made visible in a short time
  • Good questions generated for the next project meeting
  • Critical topics, delivery timelines, and open alignments identified
  • Project status and next follow-up points condensed much faster

The value was in speed, clarity, and better follow-up questions.

Appendix

From local experiment to team capability

  • Enable teams to use the workflow independently
  • Mechanical engineering environment: Google Gemini as a possible tool anchor
  • Balluff/Microsoft environment: Copilot and Azure Boards as natural starting points
  • Read-only analysis first; write paths with confirmation via proxy, MCP, and sentinel
  • Repeatable workflow: collect transcript, maintain context, read board, generate questions, improve meeting