Stop waiting on IT. | AI For You

I have been managing an AI Database product management team at Oracle for years, and I can tell you what the most common bottleneck looks like: someone on the team has a clear, specific need for a tracker, a dashboard, or a reporting template, and the answer they get is “put in a ticket and we’ll get to it in Q3.”

Claude Code is a tool that writes and runs working software from plain English instructions, and it changed that bottleneck for me completely. Not because I learned to code. I did not. But because I learned to describe what I needed well enough that Claude Code could build it, and I could have a working prototype to share with my team in the same afternoon.

This newsletter is about building the one tool your team has been waiting on, without touching a line of code yourself, without filing an IT ticket, and without waiting for engineering bandwidth that is never coming.

Here are the four moves.

Move 01: Identify the Right Tool to Build | 15 minutes

a spreadsheet flowing into automation

Claude Code is not the right tool for everything. Before you open it, spend fifteen minutes getting specific about what you need. (I skipped this step the first time and spent an hour building the wrong thing. Learn from me on that one.)

The best candidates are tools your team builds manually right now in spreadsheets or documents: the weekly status tracker someone rebuilds from scratch every Friday, the vendor renewal log that lives in a shared Excel file nobody can find, the project intake form that lands in someone’s inbox as an unformatted email and has to be re-typed into Jira, the budget reconciliation template that the finance team asks for every quarter and someone scrambles to produce.

These are not engineering problems. They are “someone needs to sit down and build this once” problems. Claude Code is that someone.

The tools that are out of scope here are anything that needs to connect to your company’s live systems, pull data from your enterprise CRM, or sit on a shared server. Those are real infrastructure projects and they belong to IT. I say this as someone who lives in the enterprise software space: the line matters, and the right corporate professional knows where it is.

Start with this framing question: “If I could hand someone a description of this tool today, and they handed me back a working version by end of business, what would I ask them to build?” Write down your answer before you open Claude Code.

Move 02: Set Up Claude Code So It Stops Slowing You Down | 20 minutes

Claude Code lives inside Claude, Anthropic’s AI platform. You do not need to be a developer to use it, but the default setup includes several permission dialogs that will slow you down to the point of frustration. Here is how to configure it so it works the way it should for a corporate professional building internal tools. (The first time I used it without doing this setup, I got asked to approve every single file change. It felt like having an assistant who needed a signature for every sticky note they placed on your desk. Twenty minutes fixes that permanently.)

You need a GitHub account. If you do not have one, create a free one at github.com. You will never write code there. It is just where Claude Code saves your project files so they do not disappear between sessions. Link it to Claude through Settings, then Connectors. Select Opus 4.6 as your model and enable “Auto accept edits.” This is the configuration that lets Claude make changes to your project without asking permission every thirty seconds.

For the fastest experience, install VS Code (free download from code.visualstudio.com) and add the Claude extension. Inside VS Code, enable “Skip Permissions” mode. This is the difference between Claude pausing to confirm every file change and Claude just building. Once this is set up, you will not touch it again.

The whole setup takes about twenty minutes the first time. After that, you open VS Code, open your project folder, and describe what you need.

One honest note on cost: if you are planning to use Claude Code daily for real work projects, the Max plan ($100 per month) is the realistic budget. The Pro plan ($20 per month) will hit token limits quickly when you are building and iterating. I cover tokens in Post 1046 if you want the full breakdown. For occasional use, Pro is fine. For building multiple tools over the course of a month, plan for Max.

Kay watching a spreadsheet rebuild itself into a clean report

Move 03: Write the Prompt That Builds Your Tool | 30 minutes

This is the move that most guides skip over, and it is the one that determines whether you walk away with a working tool or a frustrating conversation that goes nowhere.

Claude Code responds to specificity. The more precisely you describe your tool, the better the first version it returns. Here is the structure I use every time.

AI For You infographic: claude code for corporate

Part 1: What the tool does. One clear sentence. “This tool tracks vendor contracts for my team.” Not “I want something to help manage vendors.”

Part 2: What goes in. Every field, by name. “Vendor name, contract start date, contract end date, primary contact name, annual value, renewal status (Active, Pending, Expired).” If you are not specific here, Claude Code will invent fields based on what it thinks you might want, and you will spend your next thirty minutes asking it to remove things.

Part 3: What comes out. What do you see, click, or download? “A table view of all entries, sortable by renewal date, with a red flag indicator when the renewal date is within 60 days.” The output description is where most people are vague when they should be precise.

Part 4: The format. For a personal work tool, this is your choice: a single HTML file that opens in a browser tab (no server, no installation, just a file that lives in your Downloads folder) or a lightweight app that runs on your computer. I recommend the HTML file for first-time builds. It is the fastest to produce, the easiest to share with your manager as a proof of concept, and the simplest to iterate on.

Here is a full example prompt I used with my team this year:

Build me a project intake tracker for a product management team. 

Fields to capture: Project name, requesting team, business problem 
(one sentence), desired outcome, target date, priority level 
(High/Medium/Low), current status (New/In Review/Approved/On Hold).

The tool should show all entries in a table sortable by priority and 
target date. Add a "New Entry" button that opens a clean form. Add a 
status filter so I can view only entries at a specific status. Include 
a simple search bar.

Build it as a single HTML file that runs in a browser with no server 
or external libraries required. Start simple and show me the first 
version before adding anything else.

That last line matters. “Start simple and show me the first version before adding anything else” is the instruction that prevents Claude Code from building a version one that is simultaneously impressive and completely wrong. One feature at a time, confirm it works, then add the next.

Move 04: Iterate Without Starting Over | Ongoing

The first version Claude Code returns will be about 80 percent of what you need. That is the normal starting point, and what you do next determines whether you end up with a tool you put to work or a file that sits in your Downloads folder collecting dust.

Do not start a new conversation. Stay in the same session and describe specifically what needs to change. “Move the search bar above the table, not below it” is a good instruction. “Make it look better” is not. Claude Code responds to visual and functional specificity, not aesthetic preferences.

When you reference something existing, point to it. “Make the status column look like the priority column” gives Claude Code a reference to work from. “Make the table cleaner” does not.

The one thing that will consistently trip you up is what the source material calls “logic loops,” where Claude Code goes around in circles trying to fix a bug, making the same change in slightly different forms. When that happens, stop. Start a new description of the problem from scratch, describe the behavior you are seeing and the behavior you want, and send it as a fresh instruction. Nine times out of ten, the new framing breaks the loop.

a friendly terminal window

Once you have a version you are happy with, create a CLAUDE.md file in your project folder and document what you built: what the tool does, what fields it uses, what the output format is, and any design decisions you made along the way. This is the memory file that keeps Claude Code consistent when you come back to add features next week. I use one for every internal prototype I build, and it is the reason I can return to a project three weeks later without having to re-explain the entire context.

This week’s action

Open Claude Code and describe the one tracking spreadsheet your team rebuilds manually every single week. Write out the four parts: what it does, what goes in, what comes out, and what format you need. Send it. Do not wait until you have time to “learn” anything first. You already know everything you need to write that prompt.

VP @ Oracle | LinkedIn Top Voice for AI | LinkedIn Learning Instructor | Views are my own

AI For You is a weekly newsletter for corporate women (and yes, the guys reading this too, welcome) navigating careers in the AI era. Real tools, real context, no tech background required.