I‘m working also on a 1m LoC legacy system - spending much less on tokens but still achieving quality output with claude. Maybe your friend is asking the wrong questions?

GitHub - StefanKrecher/PharoCodex: Use codex app-server within Pharo Use codex app-server within Pharo. Contribute to StefanKrecher/PharoCodex development by creating an account on GitHub.

github.com/StefanKreche...
I created an interface to the OpenAI coding agent "Codex" in Pharo Smalltalk @pharoproject.bsky.social #openai #coding #software-engineering

isLoop: International School on Live Object-Oriented Programming Learn advanced object-oriented programming in Pharo

[ANN] isLoop 2026: International School on Live Object-Oriented Programming 4-6 July 2026 - Plovdiv, Bulgaria isloop.pharo.org/2026-plovdiv

AI & cost of legacy systems in UK banking - TechMotion UK banks are spending approximately £3.3 billion every year simply to keep legacy core systems running – roughly a quarter of their total IT budgets. Nearly

AI & cost of legacy systems in UK banking

I guess the challenge for legacy systems without a spec is to recreate a spec from code and runtime inspection. This should include the institutional knowledge because the effects of this knowledge must be in the code.

so true!

Tried to wrap my head around this SpecOps-method. Obviously it's crucial to have a spec when modernizing systems. Reverse engineer the spec from the code seems doable - but how can AI help to learn more about institutional/ implicit knowledge?

to anyone who is interested in legacy modernization - I created a new BlueSky feed on that topic - combining some relevant keywords: bsky.app/profile/did:...
#legacymodernization #legacycode

In some cases TDD is really helpful, e.g. complex math calculations. Put numbers in, get numbers out.

It's definitely the entrace of the mystery vortex, where Sam keeps growing and shrinking!

Despite of being more approachable, for me retro-gaming is about getting back to the way games where approached back in the day. And for a long time, softlocks have been a part of it ...
btw: MI isn't abandonware - isn't it?

Ich könnte mir vorstellen, dass es hierzu sehr konkrete Gedankenspiele gibt. Aber wahrscheinlich ist es technisch schwierig ihn zu erledigen. Und wahrscheinlich sind die möglichen Konsequenzen schwer antizipierbar ...

I‘d say it depends .. but decoupling domains might always be a good idea - even if your not following a microservices approach

to me their approach sounds like lift and shift everything slice-by-slice to the cloud - and ensuring that everything behaves the same. But the risk would be to preserve the old domainmodel and technical dept without refactoring things. Then it's running in the cloud but it's still not maintainable

doing a greenfield approach in legacy modernization of software systems is a bad idea most of the time, if you don't have a full specifiaction of the legacy system.
But talking about houses: I guess it wouldn't be a big deal if the new house wouln't look exactly like the old one

try a greenfield approach - might be easier in some cases :-)

Ain't that what characterization test are all about? AI might be pretty helpful with creating these kind of tests

I've seen a lot of projects fail at step one, since decomposition is not always that easy. When there's a domainmodel with all entities connected to each other, decomposition becomes a real PITA

this is looking really nice - can't wait to play the game!

Michael Feathers wrote in the preface of Working Effectively with Legacy Code:

“Legacy code is simply code without tests.”

But I keep wondering:

If we add tests to legacy code…
does it stop being legacy?

#legacycode #softwareengineering #softwarearchitecture

BREAKING NEWS: Actual facts appear for the first time on "Truth" Social.

I‘ve seen that pattern too - but it’s still hard to understand how to not find a job with your skill set. If your still searching- get in touch with me - there are some companies that I’m aware of that are looking for people like you

It will get worse before it gets better.

Curious where others draw that line today:
Is AI more like a smart calculator… or an external analyst who shouldn’t sit inside the red room?

Your proposal reads like a personal, modern version of that model:
AI as the “black network” — powerful, useful, but fundamentally untrusted —
and your real digital life as the “red network.”
The key isn’t zero data flow, it’s explicit, intentional boundaries.

This actually isn’t crazy at all.

In military / high-security environments this exact idea has existed for decades:
red networks (offline, trusted, sensitive work) and black networks (online, exchange, exposure), connected only via heavily controlled transfer points.

Hallo @sl-politik.de es hat funktioniert 🤩😉

So useful!

This is so cool 😍

No indie hacker ever regretted:
- Shipping before it's perfect
- Learning AWS the hard way
- Saying no to scope creep
- Walking away from the screen
- Building in public
- Open sourcing their code
- Actually finishing that side project

The ones who regret are the ones who never started.