ERNI

ERNI ERNI stands for Swiss Software Engineering to boost people and businesses in the innovation of software-based products and services on a global platform.

Physical AI works convincingly in the lab. Running it reliably on a real production line, every day, under EU regulation...
21/10/2026

Physical AI works convincingly in the lab. Running it reliably on a real production line, every day, under EU regulation and budget pressure, is a different matter. 🤖

According to Bitkom, 90% of companies see AI as decisive for competitiveness. Yet only 6% of German industrial companies already use physical AI.

What explains the gap? Rarely the model itself.

A few things worth a closer look:

● The question every team should be able to answer after a sensor or component change
● What a system should do when it meets a situation it was not trained for
● Several EU regulations converging between 2026 and 2028, and why they are an architecture decision, not paperwork
● The part of the budget that often exceeds model development

If you are planning to move physical AI from pilot to production, these are the seven factors that separate successful deployments from stalled pilots. 🏭

🔗 The full article is linked in the comments.

Most organisations have given their people access to AI tools. Emails get corrected, documents get translated, content g...
14/10/2026

Most organisations have given their people access to AI tools. Emails get corrected, documents get translated, content gets generated. 🤖

So why do the expected gains in productivity and efficiency so often fail to appear?

At ERNI, more than 50 colleagues from different teams, projects and areas have been working on exactly that question as AI Champions.

A few things worth a closer look:

● The role that sits between AI access and measurable impact
● What proved difficult when building the network
● How one recruitment team is already putting AI to work
● Five questions every AI experiment should answer before it is scaled up

If your teams are using AI but the business results are still hard to see, this is a practical place to start. 📊

🔗 The full article is linked in the comments.

Every new person on a legacy project pays a quiet tax: rediscovering what everyone before them already knew but never wr...
07/10/2026

Every new person on a legacy project pays a quiet tax: rediscovering what everyone before them already knew but never wrote down. 🧩

When a product owner left one project, that tax suddenly became very visible. It also became the starting point for LeRA, a Legacy Requirements Assistant built in a single summer.

A few things that make it worth a closer look:

● Why the part of the tool the team is most proud of never uses an LLM
● One ground rule agreed before a single line of code: LLMs cannot decide anything
● Why code and documentation are broken down into something the team calls "grunts"
● The one problem it deliberately does not claim to solve yet

If you have ever inherited a codebase and wondered how far its documentation can be trusted, this is how one team set out to answer that question. 🔍

🔗 The full article is linked in the comments.

Every technology choice in a bank or insurer comes with a bill. The real question is when you pay it, and in which curre...
30/09/2026

Every technology choice in a bank or insurer comes with a bill. The real question is when you pay it, and in which currency. 💳

C++ and Java have served financial institutions well for decades, each with its own trade-offs. Rust now offers a third option, with a bill of its own.

A few things to consider before you decide which one to pay:

● Why finance effectively runs on two different clocks
● How one banking service went from 3.8 GB of memory to 8 MB
● What caused latency spikes roughly every two minutes on a high-throughput platform, and what made them disappear ⏱️
● Where Go fits into the picture, and where it does not

The answer is not to rewrite everything. It is knowing where the current trade-off has quietly become expensive, and how to prove it before committing.

🔗 The full piece is linked in the comments.

The biggest cost of industrial software is not development. It is what happens after it ships.Industrial machines are be...
29/09/2026

The biggest cost of industrial software is not development. It is what happens after it ships.

Industrial machines are becoming more software-driven every year, and with that comes higher complexity and more responsibility when something goes wrong. A defect in one product can become a problem across hundreds or thousands of machines in the field 📉

That is where Rust is becoming increasingly interesting, not as a full rewrite, but as something that can be introduced alongside existing C and C++ code.

So where would Rust actually create measurable business value? The answer looks different for every product, and four things shape it:

● Three factors that indicate which components would pay for themselves, and which should be left exactly as they are
● Why a certified industrial safety sensor now in series production deliberately kept much of its software in C
● The 2025 figure from a major technology company comparing its Rust systems code with its own C and C++ history
● Two regulatory dates that change the economics after a machine leaves the factory, one of them already in effect

Worth reading if you are accountable for software cost, cybersecurity exposure or field service across a long product lifecycle.

🔗 Link in the comments.

The biggest cost of industrial software is not development. It is what happens after it ships.Industrial machines are be...
23/09/2026

The biggest cost of industrial software is not development. It is what happens after it ships.

Industrial machines are becoming more software-driven every year, and with that comes higher complexity and more responsibility when something goes wrong. A defect in one product can become a problem across hundreds or thousands of machines in the field 📉

That is where Rust is becoming increasingly interesting, not as a full rewrite, but as something that can be introduced alongside existing C and C++ code.
So where would Rust actually create measurable business value? The answer looks different for every product, and four things shape it:

● Three factors that indicate which components would pay for themselves, and which should be left exactly as they are
● Why a certified industrial safety sensor now in series production deliberately kept much of its software in C
● The figure Google reported in 2025 when it compared its Rust systems code with its own C and C++ history
● Two regulatory dates that change the economics after a machine leaves the factory, one of them already in effect

Worth reading if you are accountable for software cost, cybersecurity exposure or field service across a long product lifecycle.

🔗 Link in the comments.

Building an AI demo is rarely the difficult part. Moving it into production forces four questions a demo never has to an...
16/09/2026

Building an AI demo is rarely the difficult part. Moving it into production forces four questions a demo never has to answer: who is allowed to use it, what data it is allowed to see, what it costs, and what it actually did last week 🔐

When every project answers those four on its own, the results are inconsistent, expensive to build and impossible to audit.

There is another way to approach it, starting from a different premise: governance should be inherited, not rebuilt. Some of the ideas that follow from that:

● A layer that is deliberately powerless, and why that turns out to be the safest way to design it
● Data that cannot phone home, enforced at network level rather than by policy
● A code change labelled verified or unverified by the test suite, never by the agent's own report
● Four rungs of AI autonomy, from chatbot to AI-native development, and where the human sits on each one

Two remaining limitations are also named openly rather than quietly left out, which may be the most useful passage of all.

One figure to hold onto: the demo is the easy 20 percent 📊

Worth reading if you are responsible for taking AI beyond the pilot stage.

🔗 Link in the comments.

Delivery slows down. Maintenance costs climb. Quality becomes harder to protect. The roadmap keeps filling up, while con...
09/09/2026

Delivery slows down. Maintenance costs climb. Quality becomes harder to protect. The roadmap keeps filling up, while confidence in what should happen next keeps falling.

Most organisations read this as a delivery problem: engineering is too slow, the teams are not strong enough, the talent needs upgrading. Sometimes that diagnosis is correct. Often it points in entirely the wrong direction.

The uncomfortable part is that not one of the decisions behind the decline looked unreasonable at the time. Each was defensible on its own. That is exactly what makes the pattern so difficult to see 🔍

Three questions tend to expose it:

● Who owns product direction?
● How are trade-offs actually discussed?
● Which decisions are strategic, and which are simply reactions to the loudest signal in the room?

Answer them honestly and the conversation shifts. The same customer request, the same promising new technology, suddenly assessed on a completely different basis.

Worth reading if you carry responsibility for product direction, quality or cost.

🔗 Link in the comments.

Under the EU Cyber Resilience Act, a secure system that cannot be explained to an auditor is not compliant. The engineer...
02/09/2026

Under the EU Cyber Resilience Act, a secure system that cannot be explained to an auditor is not compliant. The engineering can be sound, the controls in place, and it still does not count if the reasoning behind it cannot be justified. 🔐

That single shift, from "is it secure?" to "can you prove it, across the entire lifecycle?", is quietly rewriting how connected products get built. Drawing on a real advisory programme for a next-generation connected system spanning cloud, embedded devices, wireless interfaces and third-party integrations, a few ideas worth your attention:

● Why the most valuable security work happened before the first line of code was written
● The moment fuzzing USB and Bluetooth channels revealed what conventional testing quietly misses
● Why an SBOM, Zero Trust and role-based access are no longer good practice but legal obligation
● The difference between a system that is secure and one that is defensible, and why only the second passes a notified body 📋

The Cyber Resilience Act is often read as a compliance burden. It may be more useful to read it as a change in mindset. 🤔

Link in the comments. 👇

When teams modernise a product, the conversation is all about the new platform: a new language, a new architecture, a cl...
26/08/2026

When teams modernise a product, the conversation is all about the new platform: a new language, a new architecture, a cleaner interface. One thing rarely makes it into the strategic discussion, yet it often carries the greatest risk of the entire programme. 🎯

It is the testing network. Years of accumulated tests that quietly hold together traceability, validated behaviour and the confidence that every release still works. The moment the technology stack changes, that safety net becomes exposed, and rebuilding it from scratch can cost more than rebuilding the application itself. A few things worth your attention:

● The counterintuitive decision that cut test migration effort by roughly half, and it was not about writing tests faster
● How a system of over 1,500 automated scenarios and 900 requirements moved from desktop to web without resetting product confidence
● Why the smartest move in one migration was to translate meaning rather than code, using natural language as the bridge between two stacks
● The exact role AI played, and the one thing it was never allowed to decide 🤖

The real question in any migration is not whether the system can be rebuilt. It is what should never be rebuilt at all. 🤔

Link in the comments. 👇

Adresse

Zürich
8001

Telefon

+41582681200

Website

Benachrichtigungen

Erfahren Sie als Erste/r von Neuigkeiten und Aktionen: Wir schicken Ihnen eine E-Mail, wenn ERNI etwas veröffentlicht. Ihre E-Mail-Adresse wird für keinen anderen Zweck verwendet, und Sie können sich jederzeit abmelden.

Verknüpfungen

Teilen

Kategorie