I am reading that AI techniques are increasing the efficiency of software development. It got me wondering if the Dorico team is making use of some of these new tools.
In my experience, AI tools are just as likely to rewrite your code with gibberish as they are to find/fix bugs or streamline the algorithms.
They are useful at pointing the way if youāre not an expert, but if you have a team of devs who are already familiar with the code, and good at their job, then Iād suggest the need for AI oversight is minimal.
16 Likes
Weāre beginning to use it at my day job, you donāt just hand your code off to it and it spits out some refactored version.
Thereās a lot of places it could be useful. For example, when using GPL libraries you get caught into a hell hole of EU GPDR and other regulations. Having an AI bot find and fix, say buffer overruns, would be an enormous benefit. These bugs are usually one liners, if it just found and submitted a changeset for human review that would be worth its weight.
This is nothing new, weāve had tools of varying abilities do stuff like this for years, but AI could do it much better.
For new feature development Iām skeptical with AI use. Adding features is more a problem of human judgement than writing some algorithm. The new AIās do exhibit a degree of human judgement, but if I was managing Iād rather trust my business IP on a moist human than some rando AI.
8 Likes
As someone who uses AI agents in everyday development, this is simply not true. Iāve likened AI dev tools to a chainsaw: in the hands of an expert, you can create beautiful ice sculptures. In the hands of someone who doesnāt know what theyāre doing, you can end up in the hospital.
AI as a dev tool is paradigm changing. When employed well, it works much the same way as Dorico does for us, takes care of a lot of little/simple/boilerplate things we could do ourselves, but much faster. AI doesnāt do everything, nor should it be used to. But it routinely saves me 15-30 minutes at a time. As a troubleshooting tool, it has literally saved me days.
Iām working on a mobile tuner/metronome app in my spare time. Initially, I had simple intentions for the app, and had it done with AI assistance in 2 days. But because of the power and flexibility AI has provided, Iāve added several additional features making the app a much more powerful tool to help me with my violin intonation as I learn the instrument. I never would have conceived of the additional features, had AI not taken care of the heavy lifting.
AI also improved the quality of the application. I asked the AI how my code could be optimized, expecting it to suggest various programmatic techniques. I wasnāt expecting much as code optimization is my wheelhouse. Instead, it suggested 4 signal processing algorithms Iād never heard of nor would have ever thought to have searched for. I used the agent to very quickly implement these algorithms in various combinations so I could profile them to find the best possible solution. All this took a single evening.
Just like Dorico, AI doesnāt do everything and biggest time suck is the final polish on the work.
ETA: What blew my mind about the 4 algorithm suggestions was how the agent presented them: Algorithm A could speed up pitch identification, Algorithm B could handle vibrato, etc. I didnāt give it the context of what the app did. Simply asked for code optimization ideas.
8 Likes
Not meant in any way as a blanket dismissal of AI and its good uses, but Iām just chuckling as I imagine Daniel and the team saying āyeah, we vibe-coded version 7.ā 
6 Likes
Iām curious where that āknowledgeā came from. I assume the codebase of similar apps were harvested. Is it just me who thinks the license terms of those apps might not be honoured?
4 Likes
This thought has crossed my mind as well. I wondered if Iām stealing code from other tuner apps, but realistically, how many tuning codebases are there? The knowledge is much more general than just finding like for like.
As for licensing, some repositories such as GitHub have code as fair game. As for Open Source GPL licenses, I suspect it could get interesting legally depending on how much the code matches. Code can be tricky as it is as there tends to be an ideal way of implementing a particular algorithm. (Not to mention clean-room design approaches.)
Problem in my eyes is you canāt know, and act accordingly , because AI does not deliver the context and source of that āknowledgeā. And I think thatās by design, violating or not violating enormous piles of data so itās virtual impossible to fight against it if you think your rights are violated.
3 Likes
This is the main reason anything AI-created (or substantially āhelpedā) cannot be copyrighted.
1 Like
Since Iām retired, Iām just getting started using AI in my personal software development projects. At this point, Iām not asking it to build an entire app. What Iām using it for is suggestions on how best to use a particular API to accomplish a small task. Iām still architecting things myself. Iāve seen AI make mistakes when you ask it to present an example in āimmediately useableā code - you still need to be in the loop. I am finding AI very useful for research and to provide a top level views of things. Having AI contrast my options is wonderful.
1 Like
@TonH: I have similar concerns. AI has no guardrails in terms of intellectual property rights as best I can tell (I am certainly no expert). My concern is that one personās hard work and subsequent success can be significantly diminished if the ādeciding factorā in that success is āborrowedā by AI without attribution and, in some ways, without appropriate compensation to the owner.
EDIT - further thought
What if AI platforms had to pay the intellectual copyright owner (as well as other types of copyright owners), a significant fee every time it borrowed that copyright ownerās work or creation? Now that would be an AI that I am more likely to trust.
4 Likes
Itās already eye-wateringly unprofitable as it is. This would be a death sentence. 
2 Likes
This isnāt really how AI works though. It doesnāt copy items from all the data it gathered. It uses that data to generate statistical models that it uses to create its output. Itās not much different than composers doing a lot of score study and then applying lessons learned from that to their own scores. The type of āborrowingā is more along the lines of itās a sound idea to double basses an octave below cellos, rather than copying bars of music.
2 Likes
There have been reports of text-based AI bots using copyrighted information in such things as song lyrics, high school research papers and the like. So while AI can and often does produce original information outside of text-based applications, it is known for assimilating othersā work into its own end products.
1 Like
And, that death sentence may not be a bad thing until essential guardrails are employed, validated and agreed upon. Yes, there are some benefits but I would suggest that we need protections in place before it has the green light to creep into all aspects of our digital world.
4 Likes
too late - itās everywhere already. Even if it isnāt obvious, itās there.
2 Likes
This is true. What I said is still true, albeit somewhat simplified for purposes of this thread. At the heart of the matter, AI is a tool that is neither good nor bad. There are tens of thousands of AI agents, trained in multiple ways. Adobeās agents use only licensed material, so it is possible to use AI ethically.
2 Likes
ACE studio just started AI feature to score a movie ..
Ethical? Adobe tried to get sleight-of-hand ownership of all userās creative material few years ago for AIās sake.
2 Likes