I have spent the past week in a fairly existential state, pondering what the new wave of applications will look like. This is partly fueled by the conversations I had at the AI Engineer World's Fair last week in SF, but also inspired by some recent releases from companies like PostHog, with their Slack integration, and things like Claude Tag.
Two camps
From where I stand, there are two camps: folks who live in dashboards, and others who never looked at them to begin with.
The first camp lives and breathes in the dashboards they are familiar with. This could be anything from their monitoring stack to analytics in PostHog.
The second camp is what particularly interests me: people who have never loved having more dashboards to look at but had no other alternative. Depending on your role, you have three or more to look at on any given work day. For an SRE supporting a team, this could be your logging stack, cloud provider dashboard, CI, and <insert your in-house tool>.
On the surface, what I am describing sounds solely like dashboard fatigue, which is partially true. But in the last quarter, the pace at which software is evolving is worsening this, and the human brain only has so large a context window.
Where do we go from here?
What comes next is something we have been thinking about more and more at Baseline. LLMs and agents are great at giving you more data than you need (for better or worse), but making sense of any of it would require yet another dashboard.
Two patterns I think will gain more ground in the applications of the future. The first is on-demand interfaces, where you get charts and dashboards instantly, based on your current conversation or the task you are working on.
The second is letting you bring data to where you need it. MCP is the first piece of this puzzle: you can pull information from, say, your sales CRM without leaving your terminal. Not every tool or platform offers an MCP server or an API yet, but I do think we will converge around protocols that make retrieving data much easier.
What we are doing about it
After chatting with a couple of companies that use the A2UI protocol, it became more apparent that in order for us to provide meaningful insights, we would need to switch from providing data all at once to providing data when you need it.
GraphAI is what we envision the future of cloud cost management to be: on-demand, with visualizations where relevant, metrics where relevant, and more importantly, answering the question of how much your teams, software features, and workloads are costing you.
All in one place.
Currently we are doing the "where you need it" part via our MCP server, but the next iteration of CostGraph is where we need your help.
If you are struggling with cloud spend, or trying to figure out how to cut down your compute expenses on Kubernetes, come talk to us! We'd appreciate the feedback, and you might finally get answers to the questions you have been sitting on for a while.