maker-fixer | XP | technology as a tool | knot unraveller | court jester | wandering engineer all views my own
UK
Joined March 2009
- Tweets10.2K
- Following1.6K
- Followers1.1K
- Likes4.4K
Pinned Tweet
Legacy Archipelago: a series of systems built one after another, each aimed to replace the previous one. For each step the new system only covered ~80% of the original functionality so the previous one was never decommissioned. All versions in use.
[v1]->[v2]->[v3]->[v4]->[v5]
Imagine not actually connecting critical systems (e.g. power stations, hospitals, refrigerators) to the internet.
How are people managing the org risks around LLM outages? Not just for s/w dev but for business processes. What do you do if Anthropic suffers an extended outage (days, weeks), is suddenly blocked in your region or the price goes up 10x? Seems like another us-east-1.
Gus Power retweeted
We talk endlessly about modularity in application code, then connect to a database where everything can depend on everything else. Why does modularity stop at the database?
businesssoftwareishard.com/e…
Gus Power retweeted
A chance encounter leads to an unexpected conversation about software, its history, and the questions that have been troubling us for decades.
What if some of the answers we've been looking for were there all along?
My new book, The Parnas Suite.
leanpub.com/theparnassuite
Gus Power retweeted
Why does it only fail in production?
Because production may be the first place where the (concurrent) program we wrote actually runs concurrently.
businesssoftwareishard.com/e…
Gus Power retweeted
New essay: Fifteen Hundred Classes and No Visible Shape
businesssoftwareishard.com/e…
why why why do we not have e-ink screens? please people c'mon I didn't want killer robots I just want to be able to write code outside
I wonder if LLM usage actually widens the moat - that more people just give up and say 'oh someone will just use AI to do that' and miss the changing-the-world bit?
We can expect the same system will continue to produce similar outcomes as long as the same heuristics are in use. Separate "thinkers" from "doers", silo people by function, pay bottom dollar for suppliers and staff: all of these heuristics have predictable and measurable effect.
