From 9dc1d805e006a2c1d709e2c7ae74651e58a08711 Mon Sep 17 00:00:00 2001 From: David Johnston Date: Fri, 3 Jul 2026 22:31:51 +1000 Subject: [PATCH 1/4] where are the human prompts --- ...ptss_to_stop_humans_going_of_the_rails.mdx | 32 +++++++++++++++++++ 1 file changed, 32 insertions(+) create mode 100644 src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx diff --git a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx new file mode 100644 index 0000000..f7bb8f5 --- /dev/null +++ b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx @@ -0,0 +1,32 @@ +--- +meta: + title: Where are the prompts to stop humans going off the rails? + description: Humazns are just as finicky as AI + dateCreated: 2026-07-03 + +tags: + - "ai" + +--- + +import { TextHighlight } from "@blacksheepcode/react-text-highlight"; + + +The place I work at has a miserly AI budget, and so last month I had been very disciplined in using my AI judiciously, and making use of just the Haiku model. + +Come to the end of the month, I still had more than half my budget, so I spent two hours with Opus, doing a test driven development approach to creating some custom linting rules. + +It worked really well. I copy pasted a design principles document in, and it one by one come up with example code that should fail and what the lint error should be. Tight human in the loop stuff. + +Then, I asked it to implement the lint rules, and it goes off and does. Brilliant, great job. + +Fast forward to yesterday, I think 'well Opus did a great job the other day, I've got a new task for you'. But this time I didn't do the TDD, I gave it more vauge prompts and it created _something_ that mostly worked, but I wasn't nearly as confident at it. And then I looked at my token usage that to see that I'd already used half my budget in two days. + +When we're using AI it seems that if we're not careful it starts becoming very hacky, hard coding things, copy pasting things, adding `@ts-ignores` etc, so we update our system prompts to keep it on track - "Don't do X", "if you encounter this situation, do Y" etc. + +But humans can be just as tempremental, if not more. Sometimes we're under pressure, or we're just tired, and can think 'TDD will take too long, let me just get this thing working first'* + +I think maybe it can be a good idea to include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI recieves a prompt 'we need a ClI tool to do X' instructions to have the AI come back with 'OK, are you use you want me to just go off and do this? In the past we've establised that creating a decision tree and doing TDD is is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take break?' + +Maybe I shoudn't design chatbots, because both of those sound super annoying, but that's the idea. + From 76ddec136b54d08d3f8da08f6af19f51b7eba159 Mon Sep 17 00:00:00 2001 From: David Johnston Date: Fri, 3 Jul 2026 22:37:25 +1000 Subject: [PATCH 2/4] vigalence switch --- ...where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx index f7bb8f5..6e79515 100644 --- a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx +++ b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx @@ -28,5 +28,7 @@ But humans can be just as tempremental, if not more. Sometimes we're under press I think maybe it can be a good idea to include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI recieves a prompt 'we need a ClI tool to do X' instructions to have the AI come back with 'OK, are you use you want me to just go off and do this? In the past we've establised that creating a decision tree and doing TDD is is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take break?' +There are equivilent tools in other industries - vigilance switches on trains come to mind, where the the device will make an annoying sound and the driver must press a button to prove they are still awake, otherwise the train comes to a stop. + Maybe I shoudn't design chatbots, because both of those sound super annoying, but that's the idea. From e3ec7315f1fab093cc4f222984ba76b64cf38c6e Mon Sep 17 00:00:00 2001 From: David Johnston Date: Fri, 3 Jul 2026 23:45:07 +1000 Subject: [PATCH 3/4] Rewording --- ...re_the_promptss_to_stop_humans_going_of_the_rails.mdx | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx index 6e79515..5df20f4 100644 --- a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx +++ b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx @@ -24,11 +24,12 @@ Fast forward to yesterday, I think 'well Opus did a great job the other day, I'v When we're using AI it seems that if we're not careful it starts becoming very hacky, hard coding things, copy pasting things, adding `@ts-ignores` etc, so we update our system prompts to keep it on track - "Don't do X", "if you encounter this situation, do Y" etc. -But humans can be just as tempremental, if not more. Sometimes we're under pressure, or we're just tired, and can think 'TDD will take too long, let me just get this thing working first'* +But humans can be just as tempremental, if not more. Sometimes we're under pressure, or we're just tired, and can think

To be clear, jumping straight to implmentation is often warraneted, IMO.

Particularly if you're not sure if a thing is going to work, you want a quick session to prove the concept generally, but once you start refinding the details, you're going to have a much better time driving things with clear tests.

}>'TDD will take too long, let me just get this thing working first' -I think maybe it can be a good idea to include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI recieves a prompt 'we need a ClI tool to do X' instructions to have the AI come back with 'OK, are you use you want me to just go off and do this? In the past we've establised that creating a decision tree and doing TDD is is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take break?' - -There are equivilent tools in other industries - vigilance switches on trains come to mind, where the the device will make an annoying sound and the driver must press a button to prove they are still awake, otherwise the train comes to a stop. +Maybe we can include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI recieves a prompt 'we need a ClI tool to do X' instructions to have the AI come back with 'OK, are you use you want me to just go off and do this? In the past we've establised that creating a decision tree and doing TDD is is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take break?' Maybe I shoudn't design chatbots, because both of those sound super annoying, but that's the idea. +There are equivilent tools in other industries - vigilance control devices on trains come to mind - the device will make an annoying sound and the driver must press a button to prove they are still awake, otherwise the train comes to a stop. + + From b6f185a7f91300029000ade344b6adb0c97fb36e Mon Sep 17 00:00:00 2001 From: David Johnston Date: Fri, 3 Jul 2026 23:47:48 +1000 Subject: [PATCH 4/4] Fix typos Co-Authored-By: Claude Opus 4.8 --- ..._promptss_to_stop_humans_going_of_the_rails.mdx | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx index 5df20f4..6f2ddf7 100644 --- a/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx +++ b/src/routes/posts/where_are_the_promptss_to_stop_humans_going_of_the_rails.mdx @@ -1,7 +1,7 @@ --- meta: title: Where are the prompts to stop humans going off the rails? - description: Humazns are just as finicky as AI + description: Humans are just as finicky as AI dateCreated: 2026-07-03 tags: @@ -16,20 +16,20 @@ The place I work at has a miserly AI budget, and so last month I had been very d Come to the end of the month, I still had more than half my budget, so I spent two hours with Opus, doing a test driven development approach to creating some custom linting rules. -It worked really well. I copy pasted a design principles document in, and it one by one come up with example code that should fail and what the lint error should be. Tight human in the loop stuff. +It worked really well. I copy pasted a design principles document in, and it one by one came up with example code that should fail and what the lint error should be. Tight human in the loop stuff. Then, I asked it to implement the lint rules, and it goes off and does. Brilliant, great job. -Fast forward to yesterday, I think 'well Opus did a great job the other day, I've got a new task for you'. But this time I didn't do the TDD, I gave it more vauge prompts and it created _something_ that mostly worked, but I wasn't nearly as confident at it. And then I looked at my token usage that to see that I'd already used half my budget in two days. +Fast forward to yesterday, I think 'well Opus did a great job the other day, I've got a new task for you'. But this time I didn't do the TDD, I gave it more vague prompts and it created _something_ that mostly worked, but I wasn't nearly as confident in it. And then I looked at my token usage to see that I'd already used half my budget in two days. When we're using AI it seems that if we're not careful it starts becoming very hacky, hard coding things, copy pasting things, adding `@ts-ignores` etc, so we update our system prompts to keep it on track - "Don't do X", "if you encounter this situation, do Y" etc. -But humans can be just as tempremental, if not more. Sometimes we're under pressure, or we're just tired, and can think

To be clear, jumping straight to implmentation is often warraneted, IMO.

Particularly if you're not sure if a thing is going to work, you want a quick session to prove the concept generally, but once you start refinding the details, you're going to have a much better time driving things with clear tests.

}>'TDD will take too long, let me just get this thing working first' +But humans can be just as temperamental, if not more. Sometimes we're under pressure, or we're just tired, and can think

To be clear, jumping straight to implementation is often warranted, IMO.

Particularly if you're not sure if a thing is going to work, you want a quick session to prove the concept generally, but once you start refining the details, you're going to have a much better time driving things with clear tests.

}>'TDD will take too long, let me just get this thing working first' -Maybe we can include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI recieves a prompt 'we need a ClI tool to do X' instructions to have the AI come back with 'OK, are you use you want me to just go off and do this? In the past we've establised that creating a decision tree and doing TDD is is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take break?' +Maybe we can include in our system prompts not only guard rails to stop the AI going off track, but also instructions _to guide the human_. For example if the AI receives a prompt 'we need a CLI tool to do X' instructions to have the AI come back with 'OK, are you sure you want me to just go off and do this? In the past we've established that creating a decision tree and doing TDD is most effective'. Or - 'I see that you just `git commit -am "fix" --no-verify` more than 100 files, do you think maybe we should take a break?' -Maybe I shoudn't design chatbots, because both of those sound super annoying, but that's the idea. +Maybe I shouldn't design chatbots, because both of those sound super annoying, but that's the idea. -There are equivilent tools in other industries - vigilance control devices on trains come to mind - the device will make an annoying sound and the driver must press a button to prove they are still awake, otherwise the train comes to a stop. +There are equivalent tools in other industries - vigilance control devices on trains come to mind - the device will make an annoying sound and the driver must press a button to prove they are still awake, otherwise the train comes to a stop.