My book: https://xitter.cf/t.co/QGevhw6nyE | My podcast: @HappyPathProg | @AWSCloud Agent Experience | @AgenticAIFdn TC | My opinions are mine

Crested Butte, CO, USA
Joined February 2007
Cooked up some really tasty treats over the past 3 months: - Various Jev morsels; most importantly a Scala ZIO Jev client with a full agentic loop orchestrator: github.com/jamesward/zio-typ… - Site for testing CIMD: cimd.now - Support for Mantle in ZIO Bedrock: github.com/jamesward/zio-bed… - Updates to zio-http-mcp for 2026-07-28 MCP spec: github.com/jamesward/zio-htt… - ZIO Evals lib: github.com/jamesward/zio-eva… - ZIO Git lib: github.com/jamesward/zio-git - sbt over MCP & symbol tools: github.com/jamesward/sbt-mcp - sbt-tdepver for easier transitive dep mgmt: github.com/jamesward/sbt-tde… - New Scala & Scala Native buildpacks: github.com/jamesward/buildpa… github.com/jamesward/buildpa…
Still cooking! A few more new projects in the past 2 months: - ZIO HTTP MCP client/server library: github.com/jamesward/zio-htt… - ZIO Bedrock library: github.com/jamesward/zio-bed… - sbt revolver alternative for sbt 2.0: github.com/jamesward/sbt-rel… - sbt webjars plugin: github.com/webjars/sbt-webja… - sbt plugin for sass: github.com/jamesward/sbt-sas…
1
4
27
9,486
I just noticed that elide.dev/ uses Pkl for it's build definitions - very cool! A while back I'd experimented with using Pkl for Gradle build defs and became convinced that Pkl is a fantastic option for declarative config. Glad to see Elide go this route.
10
849
AWS's @JamesWard goes beyond “hello world” to tackle real-world MCP challenges, from observability to smarter tool design. Catch him at AGNTCon + MCPCon North America as he shares practical patterns for turning basic MCP APIs into robust agent integration interfaces. Register now: bit.ly/4vkntyD
1
1
5
659
Long ago it was a developer rite of passage to create a web framework. Last year creating an agent framework was the rite of passage. This year it is agent harness. Get out there and build yourself a harness! You’ll learn a lot and might even create something exceptional.
12
1
31
2,329
JetBrains Air is significantly better than code assistant CLIs and TUIs. ACP-based harnesses w/ ACP UIs are the IDEs of the future.
15
7
1
160
19,390
Well sbt-mcp was a failed experiment. Turns out that running an MCP server for sbt, from an sbt plugin, has some problems. But from the ashes comes sbta (SBT for Agents). It's just a custom sbt launcher that adds a --mcp flag which starts a STDIO MCP server for sbt, including some useful tools like check (quick compiler - like red squiggles in your IDE) and symbol search / lookup. Check it out: github.com/jamesward/sbta
10
7
29
2,319
Yet another Agent Experience focused programming language which supports first class Effects. I think we now have about 30 of these. Great validation of Effects being critical to enabling efficient agent use! While I enjoy seeing all these new languages, I’m still sticking with Scala 3 and ZIO / Kyo as I get all the benefits of these new languages, without the maturity / ecosystem tradeoffs.
Introducing my new side project - Fai. github.com/Krzysztof-Cieslak… It’s a functional programming language with which I try to explore how we can design better “classical” software engineering tools specifically for AI agents. A couple of design choices I've made are described ⬇️
4
5
2
43
5,704
AI Agent Code Mode has become like my crazy uncle. No idea what de talkin bout.
1
11
1,406
The last moat is you. Not speed, not distribution. Anyone can build anything now. The only thing that can’t be copied is how much you care about something, what you choose to make, and what makes you you. So what do you want to make?
Distribution is not a moat. Execution is not a moat. Brand is not a moat.
7
3
31
1,922
Building a Jev-based harness. One interesting thing... You don't need a ToolSearchTool. Just let Jev decide if & what tool to use. Pretty easy optimization that has a dramatic impact on token use & latency.
9
40
4,584
Hey concurrency experts... Does this have any real chance of success?
Our goal with a scriptc is to make compiled TS competitive with go and rust. That requires robust multi-threading primitives. Notably, we are not extending JS. These programs run with the same semantics in node but in scriptc you get zero-copy immutable data structures into the worker
7
28
10,570
James Ward retweeted
I spent some "clanker" time on a port of @DHH's Campfire rewrite to Kotlin and the JVM. Surprise, attention to architecture actually matters. wecodefire.com/p/kotlin-keep…
3
10
51
3,493
We are at the point that you can imagine software exists, then it does. The next phase is when you can imagine that some else's software does something the creators didn't imagine, then it does.
4
3
2
11
2,097
One way this will manifest is that application “settings” will go away. Instead, software will be human language customizable. No need to find a setting, which may not even exist. Just tell the software how you want it to be different.
2
1
3
419
GraalVM CE compiler confidently beating C++ in inference workloads 🚀🚀🚀 By @TheMukel at #Devoxx
9
21
117
4,437
Here are the slides from my Devoxx talk on MCP 2.0 (aka 2026-07-28) jamesward.github.io/mcp2
3
15
1,264