James Ward

9.9K posts

James Ward banner
James Ward

James Ward

@JamesWard

My book: https://t.co/QGevhw6nyE | My podcast: @HappyPathProg | Working on Agentic Architectures at @AWSCloud | My opinions are mine

Crested Butte, CO, USA Katılım Şubat 2007
3.2K Takip Edilen18.1K Takipçiler
Sabitlenmiş Tweet
James Ward
James Ward@JamesWard·
What most devs get wrong about MCP is thinking that it is primarily for AI code assistants. MCP is primarily useful as an enterprise integration protocol.
English
8
11
58
9.3K
James Ward
James Ward@JamesWard·
"Developers now only review code" "Developers now only write specs" "The spec is the source of truth" "The code is the source of truth" "Code is now free" "Programming language doesn't matter" "LLMs will just write assembly" The only thing that is actually true: Everyone has an opinion but not a crystal ball.
English
7
8
117
11.7K
Jakub Chrzanowski 🥑
Jakub Chrzanowski 🥑@hszanowski·
Ok, I'm in: if you build a plugin for JetBrains IDEs, I can review your Gradle build script. I'll verify if caching works, introduce all those good practices, and even migrate it out of Groovy. Drop me your GitHub repo, and I'll pick it up next week on the live stream! On me!
English
7
7
21
2.3K
James Ward
James Ward@JamesWard·
AI Code Assistant CLIs became TUIs and are now becoming Text-based IDEs - i.e. TUIDEs. Personally, for a tool I spend all day in, I prefer GUIDEs.
English
2
2
21
2K
James Ward retweetledi
gabby
gabby@GabriellaG439·
New blog post: "A sufficiently detailed spec is code" I wrote this because I was tired of people claiming that the future of agentic coding is thoughtful specification work. As I show in the post, the reality devolves into slop pseudocode haskellforall.com/2026/03/a-suff…
English
115
266
2.5K
410.2K
James Ward retweetledi
Spring AI
Spring AI@SpringAICentral·
Spring AI 2.0.0-M3, 1.1.3 and 1.0.4 are released. 2.0 M3 brings Anthropic SDK, MCP enhancements, Jackson 3 updates, Vector Stores, ToolCallAdvisor and security improvements Find more on the releases here: spring.io/blog/2026/03/1…
English
1
16
37
2.7K
John A De Goes
John A De Goes@jdegoes·
Force your coding agent to use typed errors, so you can better prevent slop before it enters production. You say, "John, my agent doesn't want to use typed errors, because it's lazy." I say, show your agents who is boss! 😜
English
3
3
46
4.2K
James Ward
James Ward@JamesWard·
I’m working from my phone on GitHub stuff and it turns out the best workflow is to have an Agent Task create a workflow dispatch GitHub Action, then run it, and if something goes wrong, create another Agent Task to fix it.
English
0
0
4
885
James Ward
James Ward@JamesWard·
@sabine_s_ What about the whole zero value thing? please tell me OCaml doesn’t do that.
English
2
0
4
1K
sabine 🐫🦀 / 'use ocaml';
sabine 🐫🦀 / 'use ocaml';@sabine_s_·
Go isn't so bad error handling with if is mostly a cosmetic difference in terms of code readability but where Go and OCaml differ *a lot* is in how expressive the type system is (looking at Go's Nil values vs OCaml options)
Dillon Mulroy@dillon_mulroy

@WallisDev @joshua_amaju i've built very large production systems with both style of errors, both in Go and OCaml, and I very much prefer the Result type to Go's err tuple, that being said i think both are great and better than anything TypeScript has by default

English
12
4
94
14.2K
Dillon Mulroy
Dillon Mulroy@dillon_mulroy·
@WallisDev @joshua_amaju i've built very large production systems with both style of errors, both in Go and OCaml, and I very much prefer the Result type to Go's err tuple, that being said i think both are great and better than anything TypeScript has by default
English
2
0
30
16.4K
Dillon Mulroy
Dillon Mulroy@dillon_mulroy·
finally landed on a Result impl that i feel good using outside of effect code bases. neverthrow doesn't quite have the apis that i want, so i rolled my own might end up publishing it to npm
Dillon Mulroy tweet media
English
35
6
321
41.9K
Josh Long
Josh Long@starbuxman·
Happy Pi day! 🥧
English
2
1
16
2.3K
James Ward
James Ward@JamesWard·
@flixlang Have you tried the Claude Code LSP thing with Flix? Curious if that’d make things awesome.
English
1
0
1
388
flix
flix@flixlang·
Claude Code does not understand the full power of the Flix type and effect system and its complete inference:
flix tweet media
English
1
3
13
1.9K
Anh
Anh@Anh32920995·
@JamesWard I wonder which languages, frameworks are not sloppy. What if the language and framework are OK but the thing to run it is sloppy?
English
1
0
0
96
James Ward
James Ward@JamesWard·
Give slop languages and slop frameworks to an AI and you're gonna get slop code. Maybe you're ok with slop. That's ok, lots of great stuff is shipped with slop underneath. But when slop doesn't cut it, pick languages and frameworks that aren't sloppy.
English
5
4
33
2.7K
James Ward
James Ward@JamesWard·
Solving the slop PR thing is actually quite easy. Contributors must use your Skill or MCP and that thing applies your standards. In doing so it somewhere watermarks the changes. PRs are checked for the watermark and auto-closed if it doesn't exist.
English
0
0
16
3.9K
inanc
inanc@inancgumus·
@JamesWard So, you replaced CLIs with command line "interfaces"?
English
2
0
1
432
James Ward
James Ward@JamesWard·
CLIs are dead. I’ve replaced all of them with well defined interfaces exposed via standard communication mechanisms.
English
14
3
85
10.5K
James Ward
James Ward@JamesWard·
@daviddryparry x.com/JamesWard/stat…
James Ward@JamesWard

It’s absolutely wild to me that everyone tweeting about how much better CLIs or Skills or whatever is than MCP, doesn’t seem to comprehend how trivial it is to do progressive discovery of MCP tools. The Embabel unfolding tools is just one such example of easily resolving the underlying issues. Come on. These are not complex engineering tasks. It seems the AI induced brain atrophy is real. There certainly are good places for alternatives to MCP. But progressive discovery / token usage is not a factor. Most critiques of MCP boil down to the thoughtless claim that you have to include all tool metadata, automatically on every LLM call.

QME
0
1
1
103
david parry
david parry@daviddryparry·
@JamesWard Its only more tokens why not instead of having something deterministic have a script i definitely tested since i cant write a simple mcp. Better lets just execute code the LLM authors even better. Trying to be funny 😂 Maybe supervised but again determistic but then less tokens
English
1
0
1
83

Keşfet

@hszanowski @jdegoes @sabine_s_ @dillon_mulroy @WallisDev @joshua_amaju @starbuxman @flixlang