Voice of the DBA

Tag: AI

  • Claude Code Helps Analyze Test Data Manager Log Files

    I had a customer ask about analyzing their Test Data Manager (TDM) usage to determine how many people were protecting data in dev databases and how often. TDM creates log files each time it runs, but analyzing those isn’t easy.

    Or is it? Claude Code thinks it is.

    This is part of a series of experiments with AI systems.

    Consolidating Log Files

    I didn’t want to grant extra rights or permissions, and I like the loosely-coupled idea of tools, so I manually copied over log files from a couple locations into a new repo spot. This was a test to see if Claude Code could help me with something I’ve done in past with other types of files.

    TDM puts log files in Program Data, which can be locked down on many machines, so I copied things over to a location on another drive. In this case, I created a repo with a folder called TDM. In there I copied the Anonymization and Subsetter folders from Program Files. You can see the Anonymization folder contents below.

    2026-01_0212

    I opened a few random files to see what was inside and had an idea of what I’d look for in here if I were doing the analysis. I used this to help me craft a prompt.

    With that in mind, I have a good place to start, so let’s go see what my assistant, Claude, can do.

    A Quick Prompt

    I ran Claude from the TDM folder with this prompt: I have a number of log files in the subfolders here. What I am looking for is a summary of how much activity is in those log files, by month and year. The anonymization log files
    have “started” records of an “inf” type fro the classifying, the mapping, and the anonmyizing. The subsetter has “subsetter started” . Build a script that can analyze this data and
    produce summaries for the activities

    Claude agreed to help me with this response. It agrees to help summarize the data and then starts to read the files in the sub folders, looking at a sample of each. I

    2026-01_0213

    Claude then notes it understands the structure and starts to write code.

    2026-01_0214

    I have no idea how long this took as I left it alone and went to work on other things. I was finishing another piece, so a few times I approved actions to re-read files and execute the script. I know that it was less than 10 minutes, but could have been much quicker.

    When I saw the script had completed, I also had a summary. I’ll paste in the text below to make it easier to read.

    Claude’s Summary

    Perfect! I’ve created a Python script that analyzes your log files and produces a summary of activities by month and year. Here’s what the script does:

    Script Features

    The script (analyze_logs.py) analyzes:

    Anonymization logs:
    – Classify – Tracks “Classifying database” activities (task=”classify”)
    – Map – Tracks “Creating masking file” activities (task=”map”)
    – Mask – Tracks “Masking database” activities (task=”mask”)

    Subsetter logs:
    – Subsetter – Tracks “Subsetter started” activities

    Summary Results

    Your current logs show:
    – 163 total activities across both tools
    – Activity from February 2024 to February 2025
    – Peak activity in January 2025 with 69 activities (10 classify, 4 map, 8 mask, 47 subsetter runs)
    – 2025 is on track to exceed 2024 activity (88 vs 75 in the first 2 months)

    Usage

    Simply run:
    python analyze_logs.py

    The script automatically scans the Anonymization/ and Subsetter/ subdirectories and provides:
    1. Monthly breakdown of all activity types
    2. Yearly summary totals
    3. Grand totals across all time periods

    The script handles both structured JSON logs and readable text logs, making it robust against different log formats.

    Leading Me On

    At this point, I was ready to commit and save this code, but I noticed that there was a grayed out prompt already added for me at the CLI. It said: Export the summary to CSV file

    I hit Enter, and then Claude proceeded from here. I’ll show the prompt and then how Claude started to edit it’s own work.

    2026-01_0215

    Once this completed, I approved the action to run the script again and saw these results. I got a summary of the changes, and then two CSV files with the results. I hadn’t committed code yet, so in the repo, you can see the files yourself. These are in my AIExperiments repo on GitHub: TDM folder in GitHub

    I looked at the results and it looked pretty good to me. I don’t run TDM a lot on this machine, but the results seemed to match up with the dates of the log files. For example, there are log files for the anonymization (classify, map, mask) for Nov 2024, but they’re empty, so these are excluded from the csv summry.

    2026-01_0216

    What’s also neat is that the next pre-populated prompt asks if I want charts. I hadn’t thought about this, but I decided to just hit enter and see. Claude wrote a new file and then run it (I approved this) and had charts in a subfolder.

    I re-ran the TDM process today (Jan 14) to see if things worked. I got a new log file in the subsetter folder and when I ran the python script manually, I see new executions.

    2026-01_0226

    Very cool.

    Summary

    This is a quick app that is useful and it helps not only me, but the customer. In fact, lots of TDM customers and our SEs might find this useful. This is the type of data analysis I’ve done in the past, trying to look through log files and parse them, which can be a tedious task. Claude Code, (or another AI) makes this easy to generate and get working.

    Of course, this needs a bit more testing to ensure it’s accurate and I’m guessing there might be some edge cases or rough edges that need smoothing out but is that any different than an app I or someone else wrote? The app is only as good as my prompting and testing, but this was quick enough that I could either alter it myself, or let Claude do it.

    After my first Claude Code session, I had USD$1.09 left of my $5. After this, I have $0.17. I’ll buy buying more credits or subscribing as I am finding Claude to be very useful.

  • PowerPoint to HTML with Claude AI

    I had an idea for an animated view of a sales tool, and started to build this in PowerPoint. I decided to switch to Claude and ended up with an interactive one-page web app. This post looks at how I approached the process.

    This is part of a series of experiments with AI systems.

    Editing in PowerPoint

    Someone sent me a slide in PPT and asked me for feedback. I wanted to adjust the look, and started to make changes in PPT. I was moving boxes, changing a few colors, and adding something that I thought would have more impact.

    I was about to start building some animations to show how I envisioned things working when it occurred to me this would be a better as an interactive, more self-service item.

    I opened Claude and this was my prompt. I uploaded an image of my slide.

    2026-01_0265

    In 3 minutes, I had a working one-page web app that I could click around in and see changes. I had one more prompt as my gauges weren’t working correctly.

    2026-01_0271

    I took another moment to fix the UI as it had 3 of my 5 items on one line and then I had to scroll to see the others. My last prompt fixed this.

    2026-01_0272

    After getting a slide asking for feedback from a colleague, and spending 10 minutes futzing in PPT, I switched to Claude and had a working, interactive showcase of what my mind was thinking,thanks to Claude (free Claude).

    I sent it back and my colleague loves it. I don’t want to quite show this here, as we may use this in Redgate and someone will need to make it waaaaayyyy prettier than I can or even prettier than I can describe to the LLM.

    How Useful is This?

    You might say that going from an image or slide to an interactive picture isn’t useful. You might be right, and to be fair, I’ve never had the need to do this.

    However, if I think a little wider about the idea of making something interactive or expressive, I have had the need for this type of thing before.

    For example, I’ve needed a list of things that I want people to do and check them off. I’ve built this in word, with the little checkboxes, and given it to people. However, they have to remember to save as, or where it is, etc. Here, I could just give a URL to this in docs (or in code) and people could open it and live check boxes as they’re walking through a procedure. I could even have it change colors or notify when everything in a step is done.

    Or display a warning if a step is skipped. How useful would that be?

    I  have also had the need to ask a group of people a question and get their answers or ratings. How easy would it be to have a link I can send, detect the user (I’m sure AI can add this) and then record their vote, letting them override their decision at any time?

    What about a page people can use to plan vacation or anything and send me back their decisions? This saves everything in a file.

    What would your imagination let you build?

  • More Documentation is Needed

    AI is everywhere, and if you spend any amount of time looking for answers on the Internet to your coding challenges, you’ve likely encountered a lot of poor, average, good, bad, amazing, and just-helpful-enough AI content. For awhile, I was avoiding the AI summary from Google as the quality seemed slightly off, but lately it’s gotten good enough that I tend use it to decide which links to click on in the results. The summary helps me better understand the context Google sees in my search query.

    I ran across a post on coding documentation and how helpful these docs are in onboarding, code reviews, and more. The teams that worked smoothly together often had good docs that helped them function as a cohesive group. At least to some extent. Over time, teams start to depend on tools and lose some of that cohesiveness since they rely more on tools than docs. I agree with the piece that this is a part of the reason many teams don’t really function as teams over time.

    In the age of AI, this becomes more important. These AI agents are smart, but gullible and prone to making inconsistent decisions if you let them. In the piece, there’s a great quote: “When your codebase follows consistent patterns, AI assistants become force multipliers. When it doesn’t, they become chaos amplifiers.” Or as we data people know it, garbage in, garbage out.

    The lack of documentation means a lack of guidance for both humans and AI agents. It’s easy to say AI makes crazy decisions when we feed it our code, but humans do the same thing. I can’t even begin to count the number of weird decisions over structure and naming I’ve seen from other humans when I didn’t provide them guidance. It happens even when I give them standards, but at least then we can have a conversation about attention to detail if there are docs.

    I saw Brent’s predictions for AI database development in 2026, and part of the challenge in getting AI to be helpful is the lack of docs many of us have on schemas. I can’t tell you how often I’ve been asked if Redgate has tools that can doc a schema and decipher what data is being stored. Microsoft spent a billion+ on Purview, and its results in classification are a mixed bag. It’s a hard problem, and a lot of the problem is us. We don’t make good decisions about what to name columns or tables, we’re inconsistent, and we reuse columns as our requirements change, subtly altering the data being stored. Usually, this is an overloading of two types of similar, but different, data into one column. Sometimes it’s just storing whatever we want in a column (or allowing a user to do so).

    I’d like to think that the growth of AI will result in a little more attention being paid to documenting our data stores. I’d hope this results in at least using the extended properties or COMMENT capabilities of the different platforms. I think having better ER diagrams might be a second step, though certainly with some AI assistance to help keep things in sync as we evolve our schemas.

    Documentation is tedious work, and it’s not something humans are good at, or want to, update over time. However, if an AI agent were around to do the work and then let a human check the results, I suspect we might do a better job of keeping things up to date. To me, that’s another place where the AI revolution might benefit us all.

    Steve Jones

    Listen to the podcast at Libsyn, Spotify, or iTunes.

    Note, podcasts are only available for a limited time online.

  • Deep Learning and Craftsmanship Matter

    There’s concern about the future of AI and how it may affect jobs and employment for the masses. I see plenty of people on both sides of the issue. Some are sure AI technologies won’t replace people; some are concerned their jobs will be eliminated, and some are hoping that we will eliminate some jobs and create many more.

    Sometimes that’s the same person.

    A GenAI can replicate a human, or maybe more accurately, mimic one. That might work well enough for some people to trust the technology more than humans. Or maybe it works well enough, enough of the time, and lots of us are OK with well-enough. After all, I think a lot of us already work with “well-enough” or “good-enough” code on a daily basis.

    However, the GenAI is based on what other humans have already done. It’s “trained” on lots of existing ideas, decisions, codebases, etc. It can recall and use those quicker, and often as well as many humans. It might be a light craftsman, but it can’t be a great one.

    Humans will be able to deeply understand problems and create better craftsmanship for many systems. Across time, an AI can learn from these craftsman and repeat their work in other systems, but an AI will often struggle to understand the entire context of whether we would apply that solution or a slightly different one this time.

    That’s the human advantage. Deep learning and craftsmanship will differentiate us from the AIs because we can contextualize things better than an AI. Or really, we can internalize the context better than another human can express it to the AI. That will be the difficult part of working with AI LLMs, agents, and whatever comes next: explaining what is really needed in a new situation.

    Communication is hard. Because many humans aren’t good at communicating, they won’t be able to use an AI to replace other humans. They’ll struggle with the results, and they will need to hire a craftsman to help. However, that also implies that more of us need to become craftsmen, not only for the advantage it gives us over AI, but because those skills will help us better judge AI output, as well as express what we want to see the AI do.

    There will be lots of work in the future, even with AI, but I also believe that the jobs that are desired, that will pay better, will go to those who learn to use AI tech and who can judge when the quality of the work is appropriate for the situation.

    Steve Jones

    Listen to the podcast at Libsyn, Spotify, or iTunes.

    Note, podcasts are only available for a limited time online.