Timo Hahn

Business processes

8+ years in finance and sales

<New way of working>

AI Ambassador and agentic AI projects

Portrait of Timo Hahn

Timo Hahn

My mission is to close the AI adoption gap – by connecting business processes with AI expertise, an AI-first approach and a learner's mindset.

Business processes

8+ years in finance and sales: strategic planning, board decisions, ERP and planning systems, and steering 25+ markets.

Timo Hahn

Portrait of Timo Hahn

My mission is to close the AI adoption gap – by connecting business processes with AI expertise, an AI-first approach and a learner's mindset.

<New way of working>

Shaping the future of work – as AI Ambassador in a large organisation and through my own agentic AI projects.

Some of my latest work

Personal projects

Built during my sabbatical, all of them in use. Each one started with a problem I had myself. Tap any step of a pipeline to see what happens there.

AI Briefing

In daily use

~50 hSaved a year, 1 h a week

  • Docker
  • SQLite
  • NotebookLM
  • Telegram

Staying on top of AI meant reading the same news in a dozen newsletters. Maybe a quarter of each one was actually new.

Solution
A pipeline that collects newsletters, blogs and videos, scores them for relevance and turns them into a personal podcast.
How it works

YouTube channels, a newsletter inbox and RSS feeds from the big AI labs and tech blogs.

Guardrail
The agent can do exactly four tasks and declines everything else – no door left open for prompt injection.
Learning
Start with the bottleneck, not the tool. The first version was a short Google Apps Script.
Sample
Sample (in German)
Agent architecture – fetcher and analyzer run in separate containers.
Source funnel – every item grouped, rated and traceable.

Real Estate Advisor

In productive use

~50 hSaved a year, or 10× the listings in the same time

  • n8n
  • OpenClaw
  • Python
  • OpenStreetMap
  • Airtable

Finding a good investment property is a numbers game – and every listing needs its own business case and location check.

Solution
A workflow that evaluates listings on hard numbers and location quality, and keeps every result in a database.
How it works

A listing export lands in a cloud folder and triggers the workflow – no manual start needed.

Next
Growing into an AI agent for the full investment funnel – from market monitoring, business case and location check to document review, broker meeting prep and property management – plus a commercial tax tool.
Learning
Let the model judge, let code calculate. Numbers have to be deterministic.
Target architecture – the agent, its tools and its memory.
The n8n workflow, built with an AI pair developer in an agentic browser.
Code node – distances to points of interest, calculated deterministically.
Every evaluated listing lands in the database.

James, my personal AI agent

Running 24/7

~50 hSaved a year on knowledge upkeep, 1 h a week

  • OpenClaw
  • Obsidian
  • Telegram
  • Cron jobs

Knowledge from articles, projects and conversations scatters faster than anyone can organise it.

Solution
An open-source agent on its own hardware that builds and curates my second brain, and takes on smaller tasks like watching my calendar or brainstorming with me.
How it works

During the day, ideas, links and notes go to James via Telegram – from wherever I am.

Guardrail
Dedicated machine, limited tools, no access to my private accounts or credentials.
Learning
Autonomy needs guardrails: a sandbox, few tools and memory you can read.
Memory architecture – plain Markdown, no hidden state.
The wiki James maintains, as a knowledge graph.
Talking to James on Telegram.

Progressive web apps

Four apps shipped

EveningsInstead of weeks per app

  • Claude Code
  • Linux server
  • PWA

Most apps are 90 % right. The last 10 % is what annoys you every day.

Solution
Four progressive web apps, built with Claude Code on a sandboxed server – entirely from my phone, without a classic coding background.
How it works

A rented Linux server with nothing but Claude Code installed – kept apart from my private data.

Learning
Context is the product. Precise context files beat clever prompts.

Claude Code on the go

  • Smartphone interface for Claude Code on my Linux server
  • Multimodal: attachments and voice notes
  • Later replaced by Anthropic's own Dispatch feature

Meditation timer

  • Personal log-in
  • Individual settings for timer and sounds
  • Database of all saved sessions

Arcade game

  • Rebuild of a 90s Bomberman-style classic
  • Online multiplayer – join by scanning a QR code
  • Project lost. Learning: always keep backups

Rehab tracker

  • Personal log-in
  • Database for exercises, sets, appointments and a journal

Screenshot shows demo data.

New way of working – AI first

  • Agentic browser
  • Grounding
  • Context engineering

∞Time saved

Grounding

Give the model the full picture

The Real Estate Advisor worked because I grounded the model: the vision, the tools I use, and deep-research briefs for domain questions. While editing the n8n workflow, an agentic browser combined that knowledge with the live state on screen.

Context engineering

Keep context clean and separate

Working with coding agents, especially on the PWAs, showed me how much clean context matters – kept apart per app and per topic, so nothing bleeds over.

Brainstorming

Let the model ask first

Before a task, I ask the model what I've missed, or to run a Q&A with me before it starts. It surfaces perspectives I would have skipped.

Projects at work

Inside a large organisation, where the hard part is rarely the technology. Open a topic for details.

AI Ambassador

Part of the central AI network that steers adoption down to department level.

200+Colleagues trained on the AI platform
3Agent use cases scoped, implementation in progress
ResponsibilitiesSix areas, from use cases to governance
  • Identifying and supporting relevant AI use cases
  • Promoting best practices in AI adoption
  • Knowledge sharing and training
  • Collecting feedback to improve our AI journey
  • Connecting and scaling AI initiatives across sales units
  • Alignment with enterprise standards and regulation
Agent use casesInformation Architect, Campaign Analyzer, Market Expert
  • Information Architect: curates the team wiki with a human in the loop
  • Campaign Analyzer: proposes and evaluates campaigns using market context, sales figures and campaign KPIs
  • Market Expert: recurring reporting
Trainings and workshopsEnabling teams to use AI themselves

AI platform trainings for 200+ colleagues at division level, and a workshop to identify potential for AI agents in the department.

Software implementation

Multi-stakeholder collaboration: the business side of financial planning and ERP systems, from epic to go-live.

  1. Epic
  2. User stories
  3. Design review
  4. User testing
  5. Go-live
Planning tool for 25+ marketsSales, BMW

Drove a turnover planning tool from the business side via IT service providers and tenders: epic and user stories, Figma design reviews, user acceptance testing and go-live.

Business case tool for a joint ventureControlling, BMW

Moved the business case calculation into a browser-based tool: epic, user stories, click-dummy reviews, user acceptance testing and go-live.

Company-wide planning systemStrategic controlling, Personio

Part of the implementation team for Adaptive Insights (Workday): conceptualised and built the financial model, for example headcount-driven cost positions.

SAP-based cost steering systemKey user

Key user for an SAP-based cost steering system – the link between the controlling team and IT.

How I share my learnings

Three frameworks from my keynote “AI Outside-In. A Sabbatical Story”. They cycle on their own – select a layer to stay on it.

The AI value pyramid

Good results at the top make every layer below stronger.

Context is the key

Everything a model sees when it answers, in three kinds of memory.

Five levels of agentic AI

From fixed rules to full autonomy. Agentic AI starts at level three.

Higher autonomy means higher stakes – which is why every agent on this page runs in a sandbox with a short list of allowed tools.

Portrait of Timo Hahn

About me

A strong communicator

Communication and Woo – winning others over – ranked in my top strengths in 2012 and again in 2022. Consistent for more than a decade.

Fuelled by curiosity

High curiosity drives my work – especially, but not only, around AI and the future of work. Learner ranked first in 2022.

I think in outcomes

Start with the use case, not the technology. Strategic and execution-driven, with a controller's eye for the business case.

  1. German Armed Forces, mountain infantry
  2. Business studies in Hof, at Columbia University and at LMU Munich
  3. BMW, controlling – product costs and a joint venture in China
  4. Personio, strategic controlling
  5. BMW, digital sales and AI Ambassador – including a five-month AI sabbatical

Let's talk about the problem you want to solve.

Email me LinkedIn Download CV (PDF)