Our industry has moved from ‘still defining an agent’ at the start of 2025 to ‘building things that can actually build things’ by mid-year.
(That said, if you’re in the “I’d still appreciate a definition of agentic AI” stage you are absolutely not alone. This excellent explainer from Maya Murad of IBM – What You Wanted to Know About AI Agents but Were Afraid to Ask – is an excellent resource.)
Given how fast this space is moving, I wanted to highlight three recent demos that made me say “damn” when I saw them.
1. Augment Code
(demoed in Feb 2025)
Augment Code was perhaps the first agentic demo that really clicked for me. Perhaps it’s because it appealed to my former-DBA nerdiness, but their functionality that allowed users to kick off an agent to suggest updates to related SQL queries after you updated your codebase was awesome. It was my first time seeing the context window used well to span differing but related use cases.
I also really liked that CEO Scott Dietzen has an explicit vision of augmenting developers and reducing tedium rather than glibly talking about replacing them.
2. Kiro by AWS
(Participated in Summer 2025 private beta)
Amit Patel showed me an early version of Kiro and my favorite of the notes I took from that meeting was “markdown spec is holy shit!”
You can read more of my thoughts about spec-driven development here, but I from the beginning I was delighted. tl;dr I really like the process of directing an agent via specification rather than prompt.
As Andy Bell said (possibly as social media bait, but 100% spoke to me): “my favorite programming language is markdown”
my favourite programming language is markdown
3. System Initiative
(Demoed Aug 2025)
Adam Jacob and team routinely knock my socks off. The System Initiative product has evolved into an AI-native infrastructure automation platform.
System Initiative began in 2023 as a tool to help deliver on the unfulfilled dream of DevOps; their vision was to take the pain out of configuring production-ready infrastructure.
The platform creates bi-directional digital twins to help simulate infrastructure environments. In practice this means: 1. even though it has a functionality to map your system this is not a low-code tool 2. you can make changes in the system without breaking your configuration – unlike how if you made a change directly in the AWS console it would break your Terraform scripts 3. it is a “hypergraph of functions that are reactive to inputs.”
System Initiative was designed to make systems more comprehensible and easier for all developers to use before it was an AI infrastructure platform.
This latest release integrates in LLM-based workflows, so users declaratively state what they want via natural language prompt in their tool of choice (in other words, using their own LLM model and their own AI code assistant); the AI creates plans, and System Initiative simulates and executes with user approval.
System Initiative demoed four different use cases for RedMonk: understanding the infrastructure you have (without having to do any base-64 decoding), troubleshooting configuration failures, creating an automated cloud help desk for provisioning infrastructure, and migrating infrastructure across different AWS services (e.g. EC2 to ECS).
If you wanted to do this with legacy AI (I can’t believe I just typed those two words together I’m so sorry) you’d have to chain together multiple MCP servers yourself and work out transferring context across them, and then you’d still not have the benefit of testing them in a simulation environment as a safety net.
The set of System Initiative demos legit made me say “damn.” As a founder of Chef, Jacob has worked for many years on making system configuration easier for a wider audience. This demo felt like it was starting to deliver on that dream.
Disclosure: IBM and AWS are RedMonk clients.
No Comments