Category: AI

Control Lockout – An increasing danger of AI Coding

I regularly read a blog from a programmer called Simon Willison and recently he posted about something he refers to as Cognitive Debt. I prefer the term Control Lockout as it seems to me to reflect the central danger of control loss.

Control Lockout happens when organisations or individuals gradually lose understanding of systems they themselves are responsible for. The distance between those who operate the system and those who build it becomes so wide that the system becomes opaque. It still functions. It may even function brilliantly. But the understanding no longer resides inside the organisation. I personally have seen this occur in several organisations where I have worked.

Complex software stacks, outsourced platforms, and vendor-managed systems have always created this risk. What is new is the velocity. AI coding tools can generate entire architectures in minutes. That acceleration dramatically increases the risk that systems are deployed faster than they are understood.

The danger might not be that AI code is bad. Often it’s good. The danger is that it can bypass the slow, painful process through which humans build mental models. When that learning process is skipped, control migrates away from people and toward tools.

And that is the essence of Control Lockout.

If that risk is real — and I believe it is — what can be done?

1. Reconstruct Mental Models Internally

Even if AI generates the code, teams must be able to explain the system from first principles. Architecture diagrams, walkthrough sessions, internal documentation, and “explain it back” exercises should not be optional.

2. Rotate Ownership

Knowledge must be distributed. If only one developer understands the system — or worse, only the AI does — resilience collapses. Rotating ownership and cross-reviewing code forces shared understanding.

3. Enforce Explainability

If a team cannot clearly describe why something works, how data flows, or where failure points exist, that is a governance issue — not just a technical one. Explainability should be treated as a control requirement.

4. Design for Comprehension

This may be the most important principle. Systems should be structured so they can be understood. Simplicity is not aesthetic; it is strategic. Clean schemas, predictable naming, transparent logic — these are not nostalgic habits. They are safeguards against lockout.

5. Periodically Test Independence

Can the process operate without the system?
Do people know the minimum information required to perform the task?
Could the team reconstruct the workflow from first principles if required?

Simon Willison recently admitted to creating AI projects that he himself did not understand. He also wrote a blog post on it at his great blog Simon Willison Blog He also brought my attention to an academic’s experience of the very same thing Margaret-Anne Storey

PHPRunner – Pass Value to CSS Reference using Javascript in a pop up

Javascript is very powerful and will easily calculate all sorts of interesting things for you dynamically. In PHPRunner I use the popup windows for nearly every table or view form so I wanted it to work with these.

But with PHPRunner we want to store these in the database. I had a devilish time finding a way of referencing the field with which to copy any Javascript value into. After some lengthy discussion with ChatGPT 4 (via Bing) it suggested that I might try and use the CSS Selector.

I then discovered that I couldn’t seem to identify the name of the CSS Selector.

What I discovered is that I could not see a static CSS Selector reference for any of the fields except those that I had altered the formatting on for example changing the font to Roboto Mono.

So first step choose your target field and then alter it using the PHP page designer and then publish.

What I discovered was after that I could use the inspect item to identify the CSS Selector

On the published application navigate to the form and the field you wish to target for entry right click and select inspect.

Look to the DevTools window (in chrome and you should see in bold the css names of your field

You can then ask ChatGPT the following

Can you parse the input css selector I need from the following string that can be used by javascipt to be passed a value

[data-page="t0017_add"][data-itemid="integrated_edit_field4"][data-page="t0017_add"][data-itemid="integrated_edit_field4"][data-page="t0017_add"][data-itemid="integrated_edit_field4"][data-page="t0017_add"][data-itemid="integrated_edit_field4"] > * > * > input

An element is a part of a webpage. In XML and HTML, an element may contain a data item or a chunk of text or an image, or perhaps nothing. A typical element includes an opening tag with some attributes, enclosed text content, and a closing tag. Elements and tags are not the same things.

More on elements is available here

This can now be used behind a button and away we go… see below

document.querySelector('[data-page="t0017_add"][data-itemid="integrated_edit_field4"] > * > * > input').value = "Password123";

And here are some notes on finding CSS Selectors by ChatGPT4 I am still investigating CSS Selectors there seems to be a black art to understanding their structure and how they can be useful

I also asked chatGPT about child selectors in CSS. In my discussions with chatGPT I have discovered that spaces are important and symbols are important in naming and of course Javascript is case sensitive. The dynamic nature of CSS Selectors and their very specific naming conventions combined with spaces potentially being characters really means you need to be on your toes when you use them.

MS SQL Azure – Computed Column – Create UTC Datetime value from Datetime & Timezone fields – Daylight Saving

This is not tested at time of writing obtained from ChatGPT 4

Answer
Sure, here’s an example of how you might implement the ConvertToUTC function to handle time zones like ‘London/Europe’. This example uses the AT TIME ZONE clause, which is available in SQL Server 2016 and later:

CREATE FUNCTION dbo.ConvertToUTC (@localDateTime DATETIME, @timeZone NVARCHAR(50))
RETURNS DATETIME
AS
BEGIN
    -- Convert the local datetime to a datetimeoffset value using the specified time zone
    DECLARE @datetimeoffset DATETIMEOFFSET = @localDateTime AT TIME ZONE @timeZone
    -- Convert the datetimeoffset value to UTC
    DECLARE @utcDateTime DATETIME = CONVERT(DATETIME, SWITCHOFFSET(@datetimeoffset, 0))
    RETURN @utcDateTime
END

You can then use this function in a computed column definition:

ALTER TABLE YourTable
ADD UtcDateTime AS dbo.ConvertToUTC(field01, field02)

In this example, field02 should contain time zone names that are valid for the AT TIME ZONE clause. For example, ‘UTC’, ‘Pacific Standard Time’, ‘Central European Standard Time’, etc. For London, you would use ‘GMT Standard Time’.

Please note that this is just a basic example. The actual implementation would depend on how you’re storing time zone information and the specific requirements of your application. It’s also worth noting that handling date and time conversions can be complex due to factors like daylight saving time, so it’s important to thoroughly test your solution to ensure it handles all edge cases correctly.

If you’re frequently performing these types of conversions, you might want to consider storing datetime values in UTC directly in your database. This can simplify your data model and avoid potential issues with time zone conversions.

Question
Does AT TIME ZONE have a default value that will alter depending on whether the UK is under British Summer time?

Answer
Yes, the AT TIME ZONE clause in SQL Server does take daylight saving time into account. When you use a time zone name with this clause, it will automatically adjust the time based on the current daylight saving time rules for that time zone.

For example, if you use ‘GMT Standard Time’ (which is the time zone name for UK time in SQL Server), it will automatically show the correct offset for daylight saving when it is in effect.

Here’s an example:

SELECT GETDATE() AT TIME ZONE 'GMT Standard Time'

This will return the current date and time on the database BUT in my testing does NOT adjust the time to the timezone but will show the correct time zone OFFSET for the date for corrected for daylight saving i.e Greenwich Mean Time (GMT) or British Summer Time (BST), depending on the date.

Please note that the actual time zone names and their daylight saving time rules are based on the time zone data on the server where SQL Server is running. This data is updated through the Windows operating system, so it’s important to keep your servers up-to-date to ensure that the time zone rules are accurate.

And here are what I think are the supported time zones
MS Azure supported time zone values

Post-script
A lot of online editors won’t let you run the At Time Zone function but this one does..

dofactory

Comparing the time to my location and the time where I am I can tell that this is Pacific Standard Time – i.e Western America – California perhaps.

We also note that the time isn’t adjusted to the stated timezone but we do see the offset (see +01:00 in record return) Again this is because per se it doesn’t tell you what the server is set to but by using the sysdatetimeoffset we can correct the server time back to UTC and the adjust for timezone see second code example. The Timezone although included in the time is again be stated.

SELECT SYSDATETIMEOFFSET() AT TIME ZONE 'W. Europe Standard Time' as WEuropeTime, 
SYSDATETIMEOFFSET() AT TIME ZONE 'GMT Standard Time' as GMTStandard

And so there is a direction to timezone switch – In the above we have mainly been switching from UTC to a timezone but the below switches from a timezone to UTC which is what we will need if we are storing the input as a datetime and a separate timezone for each record.

SELECT SWITCHOFFSET(GetDate() AT TIME ZONE 'Pacific Standard Time', '+00:00')

Remember though timezones are held outside SQL Server databases on the server and as such are non deterministic. This is a good demonstration of determinism in practice

Deterministic algorithms are entirely predictable and always produce the same output for the same input.

Non-deterministic algorithms may produce different outputs for the same input due to random events or other factors.