The New Way of Working with Excel

The next phase of Excel work takes two new skills, and the climb only gets steeper the longer you put them off.

Looking back 3 months, I can see it now, how rarely I hand-input work in Excel. Transformation like this came on gradually, but it was sharpest most recently, while I was building a full multi-jurisdictional consolidated workbook. Part way through I realized I had not touched a single cell myself. I had described what I wanted, and something else had done the linking, the formatting, the wiring into the tabs above and below. The workbook had become a surface I was operating on through tools, one layer removed from the sheet, and the work was better for it.

If you work in Excel, working on Excel through an AI is a different act from the one you learned. You are not in the cells anymore. You are directing tools that go into the cells for you.

The closest picture I have is a machine shop. A machinist used to shape a part by hand, cranking a manual mill, pushing the cutter through the steel and feeling the resistance in their wrists. On a CNC machine they still design the part, and they still have to know it cold: the dimensions, the tolerances, what mates with what. What they no longer do is push the cutter. They hand the machine the design and the intent, and the machine does the laborious shaping. In Excel you are still responsible for the design, for understanding what you intend to build and what “done” looks like. The actual cutting into the sheet is moving to the AI tool. In Excel there are two of these tools, and they do not do the same job.

The add-in works inside the workbook

The first tool is the in-workbook add-in: Copilot, ChatGPT, or Claude sitting inside the file with its eyes on the sheet. This is for the work that happens in the cells. Integrating a new schedule. Designing a tab so it fits the workbook it is joining. Modifying a sheet you already have. Formatting.

The old way was to hand-link all of it, which meant holding the entire mechanical map of the workbook in your head: what fed a number, what pulled from it, the full path upstream and downstream. The add-in uncovers and holds that map in memory far better than you do. Ask it to integrate a new calculation and it can already see the connections and wire them in, without the mechanical grind that used to eat the afternoon.

Say you have to add an EIFEL calculation, the interest deductibility limit, to a workbook you have already built. You tell the add-in what the calculation is and where its result belongs. It reasons about the inputs it needs and the schedules it touches, because EIFEL is a known concept, already in the model’s training data, not something proprietary and private to your business.

You are not teaching it the tax. You are pointing it at your workbook.

Notice what you still have to do. You have to know, in concept, what feeds that calculation, what it does, and what flows out of it. That does not change. Your mental model of the workbook still has to be real and correct, because you are accountable and responsible for what the workbook produces at the end of the day. What you shed is the mechanical part: the hand-linking, the tracing, the clicking from tab to tab. You direct, and the add-in performs, because it holds the mechanics better than you can.

If you are genuinely unsure where something belongs, you can have the add-in lay out a plan and agree to it. Most of the time you will not need one. You state the outcome you want and let it work. You stop telling it how.

The external AI works on the workbook

The add-in stops at the edge of the workbook. The motions that run on the workbook, the automations, belong to a different tool: an external AI that writes and maintains code, a coding agent like Claude Code or Codex.

Say you have already automated the roll-forward, and you trust that process to carry the workbook from one year into the next. Add that EIFEL calculation and the roll-forward is now incomplete; it does not include the new tab. The external AI writes the new code to roll the new schedule forward, and it does not bolt that code on in isolation. The AI sees the map of your whole automation suite, how each piece connects to the next, better than you hold it yourself, the same way the add-in holds the workbook’s internal map. So it fits the new step into the existing roll-forward flow. It will even load the updated automation back into the workbook so the thing actually runs.

Where the add-in works inside the sheet, the external AI works around it. Two tools, two different territories. Keeping them straight is the practical skill, and it matters most here, because the external AI is the less familiar of the two.

Look at what all of this shares. The clicks and the keystrokes I have been describing, the linking, the formatting, the wiring of code into a workbook, are the manual part of knowledge work. There is no judgment in a column width. There is no taste in a cell reference. That is exactly the part that moves to the machine, and the judgment and the conceptual map stay with you, because that part was never manual to begin with.

This did not make me a lesser Excel expert. It made me a far more capable one.

What the next generation will not believe

My kids will not be able to fathom that people once typed code by hand. It is quite remarkable how much we built that way. Excel itself was written by hand, line by line, and that program runs the finance and operations of much of the world.

The next generation of accountants will feel the same way. They will not fathom hand-linking a workbook, or hand-formatting one: manually choosing the font and the size, dragging each column to its width by hand, typing the dollar signs in one at a time, a single underline for a subtotal and a double underline for a total, a color palette applied one cell at a time. They will find it absurd, and they will be right. It is a strange thing to add up the hours, the years really, that a profession spent fighting a spreadsheet by hand. The machine is orders of magnitude better at that manual part.

This is one step in the new way of working. It does take more than simply pointing the AI at your workbooks. Transforming your approach to work inherently requires examining how you currently perform tasks, illuminating what work gets which tool, and what might still remain manual (for now). Doing this at scale requires a vision from a higher horizon view, but doing this at any scale, large or small, requires this transitionary step.

None of this is the final skill. Using the add-in and using the external AI are two skills for the next phase, and the phase may last only a couple of years before the tools fold together and something else takes their place. That is not a reason to wait for the magic leap. Better at what, exactly? If you are not already working this way, your own learning curve only gets steeper as the tools get easier, because the gap between where you stand and where the work has moved keeps widening.

So start on the real thing in front of you. Point the add-in at the next sheet you have to build. Hand the external AI the next automation you have to extend. It is a new way of working, and I could never go back.