Field notes from Dewline engagements, organised by series. Each article is publicly readable. If you want a PDF version for offline reading or to share internally, drop your email on the article page and we'll send the one-pager.
One article per slide of our Agentic Engineering deck, written for managing directors and business units without a technical background. German and English, switchable on the page.
Article series · October 2026 · DE / EN
Prototypes now take an afternoon. Getting one into daily use is where companies stall, for two reasons that share one cause: the token bill the flat rate hides, and the eight questions enterprise IT asks that a prototype cannot answer. First article of the series Agentic Engineering, slide by slide, written for managing directors and business units.
What changes when SAP shops migrate from PI/PO to Cloud Integration, and what surfaces when Joule meets multi-country master data. Written for SAP basis leads, integration architects, and data governance owners in DACH+ setups.
Available now · 2026
One cockpit becomes four to seven tools. Payload persistence becomes opt-in. Retry stops being a first-class operation. The transparency loss isn't accidental — it's how Cloud Integration was designed.
Available now · 2026
Joule will give your CFO a confident, mathematically correct answer to a cross-country question. The answer will also be business-wrong, because field IDs were defined locally and the semantics drifted years ago.
Field note · May 2026
Section 2.2.2 walls off third-party AI agents from SAP APIs. That answers a question most SAP customers hadn't asked yet — who owns the ERP data when an agent reads it. The architectural escape hatch is a customer-owned data plane.
Field note · May 2026
Only 3% of SAP shops use Joule productively. 77% of AI-active SAP shops use Microsoft Copilot instead. That's not adoption — it's a routing decision the Mittelstand has already made about where AI workload sits in relation to SAP.
What it actually takes to move automation and AI-touched workflows from prototype to a system that runs every Tuesday at 3am without anyone watching.
Whitepaper · 2026
Most automation projects don't fail in the prototype. They fail in the slow, unglamorous distance between "it works on my laptop" and "it runs in production without supervision." This paper unpacks why that gap exists and what it takes to close it.
Topics covered:
How LLM-powered answer engines decide which brands they cite, and what it takes to be in the answer.
Whitepaper · 2026
A practitioner's view of agentic search: how LLM-powered answer engines decide which brands they cite, what structured signals actually move the needle, and how to measure presence over time. Drawing on engagements where we instrumented brands across ChatGPT, Perplexity, Gemini, and Copilot. The thesis: standalone AEO is a treadmill — AEO on a broader Growth Operating System compounds.
If there's an automation or AI topic you'd like our take on, tell us — we may already have draft notes that haven't been published yet.
Get in touch →