Agent Client Protocol
AGENTCLIENTPROTOCOL.COM
You’ve probably heard of MCP, or Model Context Protocol; it is a standard that Anthropic proposed in late 2024, that defines how AI systems (particularly LLMs) communicate with external data sources and tools. Within the context of AI-augmented coding, this allows your AI companion to do things like read API documents on the internet, or access internal datasources, querying your database for example.
Agent Client Protocol (ACP) has been proposed by the team behind the Zed editor as a standard way to plug coding agents into editors. Considering the number of coding agents that are emerging, this standard could be an important way to ensure rapid integration into your editor of choice.
It’s great to see organisations like Athropic and Zed pushing for open strandards, rather that proprietary solutions that result in lock-in.
Claude Code: Now in Beta in Zed
ZED.DEV
And hot on the heels of the ACP announcement, Zed is putting it to good use by integrating Claude Code into Zed. Previously they had their own coding agent, whereas now you have more choice.
Vibe Coding as a Coding Veteran
GITCONNECTED.COM
There are quite a lot of people sharing their experiences of AI-augmented or vibe coding, but this article stands out for a couple of reasons. The first, it’s length - a 34 minute read, but more importantly, the second reason it stands out is just how good it is.
The author describes their experiences on a fairly in-depth and hands-on exercise where they really leant into AI augmentation. They, like many of us, are in awe of what this technology can do:
“Conversations with the coding assistants are filled with sparks of what appears to be genuine intelligence that pours outside the programming box”
“You talk to these AI assistants as if they were… not machines, but incredibly knowledgeable and fast human programmers with a slightly neurodivergent mindset and a talent for sycophancy”
But this blog post isn’t just a glowing endorsement of the tools. Marco takes his time to explore the strengths and weaknesses in a very balanced fashion. He then goes beyond that to explore individual productivity, team productivity and the psychological effects of using these tools. Finally he asks whether this is a fundamental next step in the evolution of software development, from machine code, to high-level languages, functional programming … to natural language?
Are people’s bosses really making them use AI tools?
PICCALIL.COM
Software engineers are increasingly experiencing top-down mandates to use AI tools as part of their day job, with mixed results. This blog post shares a few experiences (mostly negative), from developers and designers with a range of backgrounds.
While I can understand the desire to capitalise on the promised benefits of AI, what managers and leaders need to understand is that this is a very difficult tool to adopt. AI tools are unpredictable, their capabilities are unclear, they hide their strengths and their weaknesses. They don’t even understand their own capabilities.
Furthermore, adopting these tools often requires significant changes to the way that developers and designers work.
All of this is going to take time.
So, by all means, offer up these tools to your team. But you’re not going to get the best from them unless you give your team time to learn and adapt.