Ann, Claude, Joe and the Rabbit Hole
I explained recently that Ann’s been doing a lot of work trying to organize her photographs and figuring out processes that will help her keep up with the workload of organizing, developing and uploading images in the future. If you get behind in that process, playing catch-up is really grueling. Believe me, I’ve been avoiding it for quite some time.
As you know by now, often during my work breaks I stick my head into Ann’s office and see what’s on her monitor. The real benefit (for me) of Ann’s recent efforts is that often I get to see a screen full of images, like the one below, that are simply beautiful.
Or there will be a single image that Ann will be doing her Claude thing with.
Lately, Ann has been refining a Claude agent to help her deeply analyze images, both to decide which images to further develop and to help her in the future by getting into her mind critical things to think about. Face it, Ann has pretty much gone down the rabbit hole of AI.
What is an “agent” you may ask? Well, Claude can and will break down tasks into certain types of tasks that need to be done. If you want, you can create certain “experts” that Claude can call upon to perform certain types of tasks. Claude normally creates them on the fly when multiple tasks need to be done, but if you have tasks that you know will be repeated or you want to act within certain parameters, you can create the agent and have it be the permanent expert whenever a certain type of task needs to be done. When done right, sometimes Claude does very little of the work and operates as a manager, assigning certain “tasks” to the respective expert agent(s). You could have an accountant, a graphic web designer, and, or, a physical fitness advisor. You have to spend some time with Claude to refine the agent’s skill set to have it do precisely what you want or how you want it to perform, but it usually pays off in efficiency of data collection and analysis (read: fewer tokens expended, less $ and better results). One expert said that if you think you’re going to be doing something more than three times, it’s worth the time and effort to either develop an agent or a skill for Claude to use.
So Ann has developed a photography consultant, who is much more refined than the skills that have been included in her Lightroom plug-in. Now, some folks recommend giving the agent a human name for ease of reference and . . . it makes it easier for us to remember what each agent does. It’s a lot easier to tell Claude to have Susan (the accountant agent) analyze our expenditures to see how we’re spending our money. At least, it seems more natural.
Well, Ann’s photography agent is named Joe. Gee, I wonder where she could have gotten the idea for Joe? (Cough, cough, cough, Cornish, cough, cough). Anyway, she went through a process of having Claude develop a skillset for Joe, tested it with some images, and refined the process a bit more. Ann then realized that periodically, things would seem to re-set and the comments seemed less insightful than before and she’d almost be starting over again. So he asked my help in developing a context file for Joe to consult every time Joe is called to a task. That context file was focused on what we called Ann’s photographic sensibilities, or aesthetics, so that Joe could better understand what Ann’s photographic concerns and priorities are, and what photographic matters Ann, for lack of a better phrase, could care less about (I’m over exaggerating, but you get the point). In short, Ann didn’t want a generic photographic advisor, she wanted someone who shared her photographic priorities and could help her improve in the direction she wants to go instead of imposing some generic rules of composition or photographic trends that are popular online but did not interest Ann in the least. Ann wanted a tool that understands where she is coming from, where she is going and can help her down that path instead of constantly recommending that she take another path.
My role in all this was to take a bunch of Claude conversations Ann had done (yes, she has created a Librarian agent (I don’t know it’s name) to collect and organize her different conversations so she can refer to them in the future - because Claude has no memory from one chat/task to the next), and write, in narrative form, her photographic sensibility. I wound up reading through the conversations several times and made a giant mind map of the various thoughts, preferences and issues I saw in the discussions.
Then I set about writing the context file. It came in at about 1,600 words. Ann fed it to Claude, who came back with some very interesting questions, and ultimately they clarified the statement and went about implementing it with Joe (it wasn’t a context file, but a skill . . . I think). Anyway, Ann fed a few images to Joe and the results seemed much more consistent than before. My contribution might have actually helped for once!
Ann then turned it into a new tool for her to use, and of course worked with Claude to figure out the best way to make it more efficient. Now, she just takes a screen clipping of a photograph and drops it in for Joe to analyze and she gets something like this, for the image of the reeds and hills above. It all starts with a simple statement, “Joe, take a look at this image and please give a brief review.”
As you can see, there is a real back-and-forth that, if you didn’t know better, you’d think you were chatting online with a human. And, of course, if you want to look at the image while chatting with Joe, you can click on the small icon above the initial query to check it out:
But that isn’t the only thing Ann’s been working on.
Ann has been documenting her work throughout the re-thinking of her processing methodology, in large part because much of the “processing” of images (i.e. importing the images into Lightroom, selecting them, rating them and then developing them) is only done on a periodic basis. We take a long trip, photograph a lot, and then process them once we’re back home - we only process them every few months. The rest of the time we’re developing them. Ann felt like she was doing things haphazardly from one big batch of images to another, taking a slightly different approach each time she returned from a trip. In the end, she realized everything was a mess.
So Ann’s developed a bunch of Lightroom “guides” in Word that she can now refer to when it’s time to process images after the next big shooting trip. What’s the workflow? What is the architecture of her Lightroom Collection? What do the various rating and labeling criteria (stars, colors, flags) mean? And of course, you have to have a Master Guide to let you know what the various other guides explain in greater detail.
It’s all rather impressive if you ask me - test running various scenarios with Claude to come up with a methodology for multiple tasks so you have some consistency in the future, and then documenting it for future reference.
As for myself, I’m waiting for her to complete the process because it’s all still a work in progress. Once she’s done, I’m going to check it all out and steal what I can. I could quote Picasso and say, “All artists borrow from others, great artists steal!” But I’m honest enough to admit, I’m just lazy!
So keep at it Ann! I’m benefitting from all your efforts.