X : Any ideas about dealing with legacy?
Me : Do you mean legacy IT?
X : Yes.
Me : First, I prefer to call it toxic IT given it's not something you bring the kids to see, it accumulates and can damage the health of an organisation.
X : Any advice?
Me : Well ...
Me : To begin with, we need to understand what it is. Most legacy is custom built stuff that became a commodity long ago and most (not all) legacy has been created in the last 20 years ... some is older, way older ...
Me : ... the next thing we need to understand is the fear. Everyone knows where the legacy is, most people want to remove it but there is usually real fear because if we change something then other things will break ... we often don't know how it's connected in the org ...
Me : ... lastly, we're adding to the legacy today. Everything we custom build will add to the "legacy" pile over time unless we have a means of fixing this.
X : Is there a solution?
Me : Do you mean a magic piece of software? No. Instead there is a long slog but ...
Me : ... it's doable.
X : How?
Me : Do you use test driven development (TDD)?
X : No.
Me : From now on, every new project, every upgrade, every bug fix, every change throughout the organisation must start with TDD. No, excuses.
X : How will that fix it?
Me : Within a good few years, with hard work, you'll start to build enough of a test suite that you can look to change things with a bit more confidence. You might have to commission test building for some stuff because there will be gaps.
X : Few years??
Me : Give yourself 3 to 4 years depending upon the size of the estate. Remember most of this legacy has been built over 20 years and if you've not been using TDD then you've got a lot of basic ground to re-examine. Which is also why you don't add to the problem.
X : The business won't accept that?
Me : Does it want to be more efficient, more "innovative"?
X : Yes
Me : If it doesn't fix this problem, it'll find the opposite. I did say it was toxic. Past decisions borrowed from the future, you're just paying for it now ... the bill is due.
X : And once we have the testing?
Me : Then you can start to look to tidy up things. Maps will help (you should build those as well) and patterns like strangler. Small iterative steps but start by slowly building up those tests otherwise fear will always overwhelm you.
X : Should we outsource this?
Me : To a large management consultancy?
X : Yes.
Me : No ... well, not unless you want to be beholden to them forever. You need to take control of this part. The tests are your institutional knowledge. They are your orgs brain ...
Me : ... there are some tools / groups that can help you. Search on "explainable software" by @girba
X : Calling it toxic IT is rude.
Me : It has never complained before. When you're next with your server, give it a comforting hug and just say "that Simon is a bad man".
• • •
Missing some Tweet in this thread? You can try to
force a refresh
We will be entering a phase in which the US high-tech industry (including the military complex) is highly dependent upon China, whilst China is not dependent upon the US.
For those who doubt how clear the intentions were ... go read Made in China, 2025.
China's government made its intentions evident in 2015. The US sabre rattling of sanctions reinforced that purpose whilst the US essentially continued with a misguided "market knows best" policy.
A couple of prompts with Claude 3 creates a Wardley Map for economic sovereignty in the defence space.
Not bad at all -
On par with political, military and defence folk I've spoken to. I'm also finding I can have a reasonable discussion about mapping with Claude 3.onlinewardleymaps.com/#clone:XvHskIi…
It's not perfect but it's not bad. There's more I want to interrogate Claude over ... i.e. the link to secure sourcing, the positioning of some components etc. But it's almost good enough that I can start a discussion over strategy and investment.
Anyway, upshot is that Claude 3, from my perspective, has left ChatGPT4 in the dust. Of course, I'll use Claude, ChatGPT, and Gemini to cross-compare for now but if I do start building anything more complex then the obvious path is AWS Bedrock which gives me Mistral etc.
dX: What is the single most significant problem facing AI today? Safety? Lack of skills? Inertia?
Me: Overinflated expectations by the business.
dX: You don't think AI will become widespread?
Me: Of course, it will; industrialised components are rapidly becoming cost of doing business. Don't confuse that with expectations. There will be an awful lot of disappointed businesses hoping it would create some advantage.
dX: I don't understand.
Me: Imagine you're just finishing off your plan for how AI will revolutionise your business. Six months for budget approval, one year to build team, 18 months to deliver something ... that's 3 years from now. Any advantage you thought of is long gone.
For those who don't know, I'm working increasingly on and with Glamorous Toolkit - ... I have become fascinated by our willingness to blame humans for problems that are created by our toolsets ...gtoolkit.com
... I saw this last night at Cloud Camp. Apparently, the issues with understanding, explainability and observability in AI are down to humans' inability to deal with complex environments... no, they're not. The problem is with the tools and the type of tools we are creating ...
... we've imported concepts from a physical world where tools are constrained by physics - hence a hammer is a hammer, a drill is a drill - into a world without such constraints. Rather than building contextual tools, we've built constrained tools.
Faulty products, harm to users, executives profiteering, fighting compensation ... what is truly bizarre about the Fujitsu Horizon case is that the public seems to think that this is an isolated example rather than the normal way that traditional corporations act.
Just take a look into any industry, pick something like retail with BNPL (by now pay later) to EWA (earned wages access) to use of slave labour. It's story after story of despicable behaviour, of exploitation of both workers and consumers in pursuit of profit.
Or pick something like energy, where misinformation and self-interest abound from carbon capture to hydrogen - both technologies which are not primarily for the benefit of consumers or the environment but instead prolong a fossil fuel industry and all the harm it causes.