AI seems to be part of every conversation in the software world these days, and quite a few conversations outside of it. The technology is moving fast, and companies everywhere are trying to figure out what it means for their business and how they can put it to work. We have been asking those same questions about Dodeca.
We decided to play a bit of a wait-and-see game because we wanted to understand where AI could provide real value for our customers and how it could best work with the spreadsheets they use every day. We have now put quite a bit of infrastructure in place, and the first visible pieces are coming together. I thought this would be a good time to explain our approach, show what we have been working on, and share where we think this is going.
One prediction I keep hearing is that AI is going to kill spreadsheets. That sounds familiar. When I started this company in 1998, I was told that spreadsheets would go away within two years because the web would take over and nobody would ever use a spreadsheet again. Well, here we are in 2026, and spreadsheets are still very much with us. Now we are hearing the same prediction about AI, and I still don’t believe it.
A spreadsheet is a natural interface for viewing and working with tabular data. Rows, columns, calculations, and the ability to see numbers in context remain useful regardless of where the data comes from or how it was produced. I expect spreadsheets to be a very useful way to display and explore the results of AI, and that thinking is an important part of our approach to AI in Dodeca.
Our first step was to build the infrastructure for AI connections. Dodeca has a modular architecture, and we have extended that architecture so AI connectors can be developed and added as modules. We have added modules for connecting to ChatGPT, Anthropic, and Microsoft Copilot, with others on the way.
Why take this approach? Companies are making their own investments in AI. They are selecting providers, establishing security requirements, and deciding how AI fits into their business. Some are using multiple providers, and others may change providers as their needs evolve. We want Dodeca to work with those decisions. You should be able to connect Dodeca to the AI your company has chosen, and our connector architecture gives us a foundation for supporting those choices as the technology changes.
Of course, having a connection is only the beginning. You need to do something useful with it. We have added scripting functionality that allows applications to use generative AI to add content to worksheets. We have also created an AI chat pane that can be plugged into Dodeca as a module and connected to your selected AI provider.
The chat pane is probably the most visible part of this initial work, but what happens under the covers is just as important. Getting a useful answer from AI depends quite a bit on what you give it. You need to ask the right question, provide enough detail, and supply the right context. Those of us who have been around computers for a while know the expression “garbage in, garbage out.” That still applies.
Dodeca already has a lot of the context a user would otherwise need to explain. The chat pane automatically packages the spreadsheet in scope and sends it to the AI along with the question. It also includes information about the connections being used by that Dodeca view, providing context about the underlying data sources. With the appropriate tools configured, the AI can use that information to look for additional detail.
You can also configure custom context to be passed along automatically, such as business terminology, background information, or instructions relevant to your application. The idea is to let users ask questions about the report they are looking at without having to describe the entire report first. Dodeca can supply that context for them, giving the AI a better opportunity to return a useful answer.
Much of the discussion about AI in software focuses on how AI adds value to a product. We certainly see opportunities for that in Dodeca, but we also see an opportunity for Dodeca to help people evaluate the answers they get from AI. One of the things we plan to work on is a Dodeca MCP server. MCP, or Model Context Protocol, provides a way for AI applications to work with external tools and data. Our vision is for an AI application to use Dodeca to help identify reports that can validate the results returned from an AI query.
Imagine asking an AI application about a variance and receiving an explanation. A Dodeca integration could suggest a relevant report that lets you check the explanation against a vetted, curated report running known application logic. That would give you a practical way to compare the AI’s answer with established reporting and identify possible errors or hallucinations.
The report could also provide something beyond verification: more context. An AI response might identify a number or a possible cause of a variance. A report could show the breakdown by product, department, or period, along with the surrounding numbers. Seeing that broader picture may help you understand something that is difficult to capture in a chat response alone. The report can help you check the answer while also giving you a better understanding of what the answer means.
The video below shows our initial AI chat functionality in Dodeca, using sample data from our friends at Caprus AI. Dodeca’s AI chat connection can be configured to use one or more MCP servers. In this example, we are using the Caprus MCP server connected to Essbase to explore the data underneath the report and bring back additional insights.
We will continue to expand the chat pane’s functionality, and the planned Dodeca MCP server will take us further toward using trusted reports to help evaluate AI answers. For now, the video should give you a good idea of what we have working today. Enjoy!
