3
points
ColinEberhardt
Joined 1,107 karma
- Ironic, yet practical. This is so that the issue can be 'pinned' to the top of the Issues page, making it more visible:
- Nolan shares some interesting reflections at the end of this post. Here are a few direct quotes:
> "I’m somewhat horrified by how easily this tool can reproduce what took me 20-odd years to learn"
> "I’ve decided that my role is not to try to resist the overwhelming onslaught of this technology, but instead to just witness and document how it’s shaking up my worldview and my corner of the industry"
> "I have no idea what coding will look like next year (2026), but I know how my wife will be planning our next vacation."
- It’s still around, and tends to be adopted by big enterprises. It’s generally a decent product, but is facing a lot of equally powerful competition and is very expensive.
- “When an LLM can generate a working high-quality implementation in a single try, that is called one-shotting. This is the most efficient form of LLM programming.”
This is a good article, but misses one of the most important advances this year - the agentic loop.
There are always going to be limits to how much code a model can one-shot. Give it the ability to verify its changes and iterate, massively increase its ability to write sizeable chunks of verified and working code.
- 1 point
- Oh, so _that_ is what a sub-agent is. I have been wondering about that for a while now!
- Likewise. I have a nasty feeling that most AI agent deployments happen with nothing more than some cursory manual testing. Going with the ‘vibes’ (to coin an over used term in the industry).
- > We find testing and evals to be the hardest problem here …
I wonder what this means for the agents that people are deploying into production? Are they tested at all? Or just manual ad-hoc testing?
Sounds risky!
- Neat! Really like this, and will be interesting to see how it tracks over time.
My only concern is the scale you have selected as we are already close to 100, are you going to have to allow it to be ‘turned up to 11’ at some point?
- 3 points
- 4 points
- As an aside, the Ghostty recently made it mandatory to disclose the use of AI coding tools:
- Thanks Simon - you asked us to share patterns that work. Coincidentally I just finished writing up this post:
https://blog.scottlogic.com/2025/10/06/delegating-grunt-work...
Using AI Agents to implement UI automation tests - a task that I have always found time-consuming and generally frustrating!
- 1 point
- 1 point
- I agree with the overall sentiment here, having written something similar recently:
“LLMs don’t know what they don’t know” https://blog.scottlogic.com/2025/03/06/llms-dont-know-what-t...
But I wouldn’t say it is the only problem with this technology! Rather, it is a subtle issue that most users don’t understand
- I had a car from the same era, the Boomerang, similar in style but 4WD:
https://randomcompetitions.co.uk/wp-content/uploads/2023/12/...
Absolutely loved that car, used it for hours and hours every week. The best 'toy' I ever had (other than my Amiga A500!).
- Oh wow, so cool to meet someone who used an app I wrote 25 years ago. Great to hear from you and congrats on the prize!
- Very cool. I actually created a shareware application that rendered Photomosaics 25 years ago! Here is a link to "Mosaic Magic" from the Wayback Engine:
https://web.archive.org/web/20010405175706/http://fishsoft.c...
I managed to make a decent amount of spending money from that application during my University years.
One interesting lesson I learnt was about enterprise pricing. I think I charged ~$20 for Mosaic Magic, however, I started to get emails from organisations asking about pricing for commercial use. Nothing on my pricing page suggested that commercial use was prohibited, I guess they just thought $20 was rather cheap. From there-on, I charged $150 for "commercial use". Basically, anyone who thought they should be paying more, did!
Finally, Robert Silvers filed a patent and trademark for Photomosaics in ~2000, and was, for a while aggressively pursuing organisations that he felt was infringing. I assume this has all died down now.
- That is very cool, thanks for sharing, and congratulations on an epic streak.
I'm also into running visualisation, and created the running report card:
It visualises your year in running, with some fun narrative generated by GPT. Here's my report card:
- TL;DR; they report that AI tools slow down task completion by 19%
- 18 points
- until some of the significant flaws of agents are addressed (hallucination, explainability, bias), I'm not really all that interested in extending this model further.
Agentic AI definitely works for software engineering because we have suitable mitigations for its limitations. It is unclear what those mitigations might be in other fields of application.
- There are so many prompting guides at the moment. Personally I think they are quite unnecessary. If you take the time to use these tools, build familiarity with them and the way they work, the prompt you should use becomes quite obvious.
- Thanks for sharing pmabanugo, a couple of those posts are new to me too. If you’re taking submissions, I’ve been exploring how to make the most of these tools for the past few months, here’s my latest post:
https://blog.scottlogic.com/2025/05/08/new-tools-new-flow-th...
- Great post, thanks for sharing. I wrote something similar a couple of years ago, showing just how simple it is to work with LLMs directly rather than through LangChain, adding tool use etc …
https://blog.scottlogic.com/2023/05/04/langchain-mini.html
It is of course quite out of date now as LLMs have native tool use APIs.
However, it proves a similar point to yours, in most applications 99% of the power is within the LLM. The rest is often just simple plumbing.
- Agreed, embeddings are a very useful tool. Started using them for article recommendations on our blog a few years ago, with a measured increase in engagement:
https://blog.scottlogic.com/2022/02/23/word-embedding-recomm...
- 1 point
https://github.com/nibzard/awesome-agentic-patterns/commits/...
Unfortunately it isn’t possible to detect whether AI was being used in an assistive fashion, or whether it was the primary author.
Regardless, a skim read of the content reveals it to be quite sloppy!