AI Transformation

My Thoughts on AI Transformation

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  5d ·

  If AI makes implementation cheaper, engineering leadership does not become less important.

  It becomes more important because the bottleneck moves.

  When writing code is slow, weak judgment can stay hidden for a while. When every engineer can produce more output, weak judgment becomes visible much faster.

  The leadership question shifts from “can we build this?” to:

  • should we build this?
  • is the requirement clear enough?
  • where did the bottleneck move?
  • what needs review?
  • which AI-assisted workflow should become shared practice?
  • where are people over-trusting or under-using the tools?

  AI adoption in engineering is not just a developer productivity topic.

  It changes discovery, review, testing, onboarding, incident response, documentation, and the way teams learn.

  That needs leadership.

  Not command-and-control leadership.

  Leadership that makes private experiments visible, turns useful workflows into shared capability, and creates enough safety for people to say two things honestly:

  “I used AI here, and this is how I checked it.” “AI was not useful for this task.”

  Without that honesty, leaders get adoption metrics and lose operational truth.

  AI accelerates implementation.

  Leaders shape the environment where implementation becomes valuable.

  #EngineeringLeadership#EnterpriseAI#AITransformation

  

  sebastianstoehr.de

  The New Role of Engineering Leaders in AI Transformation

  Why cheap implementation makes engineering judgement more valuable, and why leaders must turn private AI experiments into shared team capability.

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  1w ·

  Most companies do not start their AI journey with an operating model. They start with a mess: ChatGPT tabs, Copilot habits, Claude research workflows, hidden scripts, and prompts nobody owns.

  I do not think that first mess is the problem.

  It is often the first honest signal that people are trying to make AI useful in their actual work. None of that usually starts in a steering committee.

  The problem starts when the mess becomes invisible infrastructure.

  • Sensitive data goes into tools nobody cleared.
  • Three teams build the same workflow separately.
  • Quality standards differ from team to team.
  • A useful experiment quietly becomes business-critical.
  • Nobody can explain who owns it.

  The usual reaction is policy theater: approvals, boards, templates, and procurement gates for every small experiment.

  That does not create control. It pushes the work underground.

  The better path is an Enterprise AI toolbox.

  Not one assistant for everyone. Not chaos forever.

  A practical toolbox separates the layers: everyday assistants, team workflows, shared knowledge access, production agents, and reusable assets.

  The core move is to let AI work graduate.

  Exploration should stay lightweight. Pilots should document users, data, risks, and review. Managed workflows need quality gates and owners. Production services need the same discipline as any other business-critical system.

  Too little control creates risk and duplication. Too much control kills the learning you need to manage the risk.

  I wrote the full article on moving from AI tool chaos to an Enterprise AI toolbox without killing the curiosity that created the first useful use cases.

  Link in the comments.

  What is the biggest challenge you are facing when moving AI from scattered experiments to something the organization can actually rely on?

  #EnterpriseAI#AITransformation

  → From AI Tool Chaos to an Enterprise AI Toolbox

  Stop Turning AI Chaos into policy theater

  • 
  • 
  • 
  • 
  • 
  • 
  • 
  • 
  • 
• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  2w ·

  Enterprise AI agents do not fail because they cannot generate text.

  They fail when they do not understand the context.

  A few weeks ago I wrote about the “Confluence Graveyard”: the place where important knowledge slowly gets lost across old pages, tickets and documents. Since then, I made great progress: The knowledge base is now kept up to date regularly, currently twice a day, by consuming relevant tickets, pages and documents. It is already being used to enrich agentic workflows — for example to interpret numbers with more context and create better reports.

  Google recently introduced the Open Knowledge Format, which specifies best practices to share knowledge for agentic use cases. They kept it super simple:

  • Markdown files
  • YAML metadata
  • Normal links
  • Versionable structure
  • Readable by humans and usable by agents

  And importantly: it does not force a fixed domain structure.

  That makes it very compatible with real enterprise work, where each organization needs to model its own products, metrics, processes and decision history.

  For me, this confirms an important direction:

  The next step for enterprise AI is not only better prompts. It is better shared context.

  
• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  3w ·

  The best first AI agent for your company is boring. On purpose.

  We’ve all seen the demo: the agent plans a project, browses the web, writes the code, updates three systems. Genuinely impressive. But it skips every question that actually decides whether an agent ships: What is it allowed to touch? What can it change? Who’s accountable when it’s wrong? How do you even know it’s wrong? In a keynote you can wave those away. In a real company, they are the project.

  So my advice is almost annoyingly modest. Don’t start with the agent that “does my job.” Start with a narrow workflow that already happens every day, has clear inputs and outputs, can be checked in seconds, and fails in ways you can recover from.

  Two tests decide it: is it bounded, and is it reviewable?

  Drafting release notes from merged pull requests passes both. So does classifying incoming tickets against a taxonomy you already use. Nobody puts these on a conference slide. They also work.

  Boring isn’t the consolation prize here. It’s the strategy. A bounded agent that works the fiftieth time builds more trust than a flashy one that fails once. And trust is the thing you actually need before you earn the right to do anything ambitious.

  Autonomy should be earned through evidence, not assumed because the tech is exciting.

  🦾 Which boring workflow would your team automate first?

  #EnterpriseAI#AIAgents#AITransformation#AIStrategy

  

  sebastianstoehr.de

  Start With Boring Workflows

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  3w ·

  Most enterprise AI initiatives are stuck in a loop. One impressive PoC after another gets built, applauded, and then quietly forgotten.

  Why? Because we confuse technical feasibility with business readiness.

  Building a prompt that works once in a controlled showcase is easy. Building a reliable workflow that survives the daily mess of complex edge cases, compliance constraints, and time pressure requires real product ownership.

  Before you fund or start your next AI prototype, run it through this 5-Question Gatekeeper:

  • The Task: What specific, painful task is this solving, and how are users doing it today?
  • The User: Who exactly is the target user, and what established habit must they change to adopt this tool?
  • The Quality: What defines “good enough” for the output, and what is the exact tolerable error rate?
  • The Workflow: What is the human’s exact role? Do they review, edit, approve, or just monitor?
  • The Lifecycle: Who owns the maintenance of prompts, context data, and automated tests three months after the demo is over?

  If nobody takes ownership of these five questions before development begins, you are building a demo, not a product.

  Stop optimizing for demo applause. Start optimizing for users who return to the tool week after week.

  #EnterpriseAI#AIStrategy#AITransformation

  

  sebastianstoehr.de

  Escape the AI PoC Trap

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  1mo ·

  Ever feel like modifying AI prompts is just an endless cycle of fixing one thing and breaking another?

  You tweak a sentence to improve a specific use case, only to find that you’ve completely broken the use case you added last week. In the AI world, prompts, system instructions, and agent workflows define software behavior — and behavior can regress. Tracking these side effects with manual reviews alone is virtually impossible.

  If you want to transition from experimenting with AI to using it professionally, reliability is essential.

  🎯 The solution? We need to shift our mindset and start treating our AI setups like traditional software.

  In this article, I share the concept behind building an automated AI test suite and introduce tools like #promptfoo. It’s the first step to relying more on facts than your gut feeling alone.

  How are you currently managing and testing your AI workflows?

  #AIStrategy#EnterpriseAI

  

  sebastianstoehr.de

  Your AI setup needs tests, not just intuition

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  2mo ·

  We’ve all been there: Information buried in a chat thread from 2022, an outdated Confluence page, and three different JIRA tickets.

  I’m currently working on a pipeline to turn this “raw data swamp” into a Domain-Specific Knowledge Graph. Here is the 2-step approach:

  Step 1: The Archaeologist (Extraction)

  Using MCP to connect to our systems, this skill:

  • Merges historical data into a “current truth” markdown file.
  • Maintains a changelog of how a topic evolved.
  • Links every claim back to the original source.

  Step 2: The Librarian (Interlinking)

  A folder of files is just a library without an index. The Librarian:

  • Identifies relationships (e.g., “Product A is an add-on to Product B”).
  • Creates a navigable “Markdown Wiki” for AI.
  • Enables complex reasoning (e.g., explaining sales trends based on product dependencies).

  How are you ensuring your AI has the “current truth” and not just some (maybe outdated) data?

  

  sebastianstoehr.de

  How to Turn Scattered Data into AI Intelligence

• 

  Sebastian Stöhr

  Turning Enterprise AI from pilots into practice | Sr. Advisor AI Transformation

  2mo ·

  24 Hours. One Production-Ready iOS App. 🚀

  By leveraging modern tools, I integrated AI meal scanning, Apple Watch support, and Home Screen widgets into a diet tracker in record time.

  This experiment proves that the barrier between “idea” and “execution” has vanished. We need to stop optimizing old workflows and start reinventing how product teams actually work together.

  

  sebastianstoehr.de

  Building an iOS app in just a day