What is Agentic AI? A Photographer's Guide to Automated Workflows

WEBVTT
00:00:01.082 --> 00:00:06.750
What if the next useful thing for photographers isn't another app you open and close?
00:00:07.708 --> 00:00:13.208
What if it's a system that actually does part of the work while you're not even looking at it?
00:00:14.541 --> 00:00:17.625
That's what Agentic AI actually means.
00:00:18.166 --> 00:00:25.375
And it's worth 20 minutes of your time because it's showing up in your workflow, whether you invite it in or not.
00:00:25.625 --> 00:00:45.875
(Music) One definition, one workflow example, one decision framework you can actually use no tricks, no doom and gloom.
00:00:46.165 --> 00:00:52.457
Just what this is, where it fits and how you decide if it's worth your time.
00:00:53.790 --> 00:00:57.415
And to be clear, this isn't some future concept.
00:00:58.375 --> 00:01:05.957
Tools that flag instead of decide are already shipping in mainstream photo software right now.
00:01:06.582 --> 00:01:10.875
Let's start with the definition because this word gets thrown around loosely.
00:01:11.707 --> 00:01:15.332
I want you to walk away with a better understanding of it.
00:01:16.457 --> 00:01:26.457
But real quick, before we get into this, I've been a professional photographer since 2016 with two decades in IT behind that.
00:01:27.250 --> 00:01:32.457
So that's where I'm coming from inside of both worlds.
00:01:33.790 --> 00:01:34.957
Here's how I see it.
00:01:35.707 --> 00:01:59.500
Most AI tools you've already used are like a fast intern at your desk who only moves when you tell them to. You ask one question, you get one answer and that's it until you ask again, write this caption, answer this email, enhance this photo, one ask one output back to square one.
00:02:00.915 --> 00:02:02.707
Agentic AI is different.
00:02:03.875 --> 00:02:07.832
Think of it like a second shooter you've briefed for the day.
00:02:08.915 --> 00:02:19.875
In the morning you set the goal, cover the keynote, get the breakout sessions, work the sponsor booth during lunch and flag anything unusable.
00:02:20.915 --> 00:02:24.790
You're not standing over their shoulders for every single image.
00:02:25.625 --> 00:02:26.832
They go do the job.
00:02:27.332 --> 00:02:29.457
They make small decisions along the way.
00:02:30.290 --> 00:02:33.540
They check in with you when it actually matters.
00:02:34.582 --> 00:02:36.125
That's the shift right there.
00:02:36.750 --> 00:02:39.540
It's not one question, one answer anymore.
00:02:40.082 --> 00:02:47.875
It's one goal and a string of actions taken on your behalf with less of you in the loop at every step.
00:02:49.332 --> 00:02:53.040
So in your workflow, it looks like this.
00:02:54.040 --> 00:03:06.457
Cull this folder, flag the keepers, draft a note to the client, hold a time on the calendar, one instruction instead of four you normally have to trigger yourself.
00:03:07.750 --> 00:03:09.750
Let me make that real for a second.
00:03:11.207 --> 00:03:23.500
Say you just wrapped up a day long corporate conference, three keynotes, four breakout sessions, headshots at the sponsor booth during lunch and the after hours networking reception.
00:03:23.500 --> 00:03:29.040
You come home with 2,000 images and you already know what happens next.
00:03:30.207 --> 00:03:32.250
It's not just find the sharp ones.
00:03:33.000 --> 00:03:34.875
You've got CEO moments.
00:03:34.915 --> 00:03:59.082
The client needs for the recap deck, the logo shots, the sponsor requested, the engagement shots, the events team wants for social, the candid networking images for the internal newsletter, four different deliverables, four different sets of criteria, all sitting in the same folder.
00:03:59.957 --> 00:04:19.875
Now, imagine you gave a system one instruction, sort the shoot by deliverable, flag the keepers in each category, route the finals into the right client folders, draft your gallery is ready note and hold Tuesday afternoon for review call.
00:04:21.333 --> 00:04:27.290
And, it did all that while you got some sleep! That's the shape of agentic.
00:04:27.708 --> 00:04:36.250
Whether any tool actually delivers that reliably today, that's a separate question. That's where we're headed next.
00:04:38.125 --> 00:04:44.915
But, there is one thing worth noting before I go further, because this matters more than the definition itself.
00:04:45.790 --> 00:04:49.625
Agentic is a fast-moving label.
00:04:50.540 --> 00:04:56.583
What a vendor means by it this month might not be what they mean by it next quarter.
00:04:56.583 --> 00:05:02.290
So, treat everything I say here as snapshot not the final word.
00:05:02.708 --> 00:05:10.750
I'm recording this today if you're catching this six months from now. Go check whether things have moved.
00:05:11.708 --> 00:05:12.958
They probably have.
00:05:13.750 --> 00:05:19.540
Which is why I want to level with you on how I'm calling things today.
00:05:21.000 --> 00:05:24.708
Everything I say falls into one of three buckets.
00:05:25.250 --> 00:05:28.750
Confirmed. Likely. Or just my Opinion.
00:05:29.415 --> 00:05:40.250
I'll try to make it obvious which one you're getting because, I'd rather you double check me than trust me blind and get burned later.
00:05:41.540 --> 00:05:45.290
So where does this actually touch your work?
00:05:46.290 --> 00:05:48.500
Forget whether AI is impressive.
00:05:48.790 --> 00:05:49.915
That's not the question.
00:05:50.375 --> 00:05:53.583
The question is whether it fits tasks.
00:05:53.958 --> 00:05:54.958
You already do.
00:05:56.040 --> 00:06:02.125
Like culling, client communication, scheduling, file routing, follow-up.
00:06:02.125 --> 00:06:12.833
Those are the categories worth testing because they're repetitive. They eat your time and they don't always need your full creative judgment at every single step.
00:06:14.000 --> 00:06:28.458
Let me paint the picture and I want to be straight with you here I'm not telling you one tool nails all five of these steps today. I haven't verified that so I'm not selling it to you as fact.
00:06:28.790 --> 00:06:36.165
Think of this as a mental model. Use it to test any tool that claims to do this.
00:06:36.165 --> 00:07:07.415
Alright, five steps start to finish let's walk through them together First one intake photos land from the shoot and an agentic system scans the batch groups by exposure focus and duplicates Culling support it flags likely keepers and likely rejects. Flags not decides that distinction is basically the whole episode in two words.
00:07:09.375 --> 00:07:30.000
Routing once you've got your set it routes files backups maybe delivery based on rules you've already set Client communications. A draft message goes out or it waits in your queue for thumbs up your gallery is ready or something like that.
00:07:31.583 --> 00:07:37.375
Scheduling Got a follow-up call or a print consult on deck.
00:07:37.915 --> 00:07:43.000
It proposes calendar holds around when you're actually free.
00:07:44.000 --> 00:07:53.208
Now before you go getting all excited about all of that. Think about how any of those five steps could go sideways.
00:07:54.540 --> 00:08:05.250
Here's the scenario that worries me that same corporate shoot. The system flags 400 of your images as likely rejects.
00:08:06.333 --> 00:08:13.083
Auto moves them to a trash folder and most of those calls were probably right.
00:08:13.083 --> 00:08:31.708
But some of them were the ones you would have flagged as keepers the image where the CEO finally laughed instead of holding her boardroom face. A moment a sponsor rep and their top client actually connected at the reception.
00:08:32.666 --> 00:08:40.750
The candid where the event lead saw all of her planning pay off. The system doesn't know those moments matter.
00:08:41.250 --> 00:08:46.750
It's grading on exposure and focus not story.
00:08:47.250 --> 00:09:06.332
That's what overreach looks like in real terms. Not a system that fails a system that makes the right technical call and the wrong storytelling one and you don't catch it unless a human is still in the loop between flagged and deleted.
00:09:07.291 --> 00:09:08.166
Quick pause here.
00:09:08.250 --> 00:09:09.582
This will only take a second.
00:09:10.082 --> 00:09:16.207
If you're one of my audio only listeners, know that this episode is also up on my YouTube channel.
00:09:16.416 --> 00:09:19.207
At @ShutterandAlgorithms, all one word.
00:09:19.625 --> 00:09:22.707
If seeing the workflow breakdown laid out visually would help.
00:09:22.957 --> 00:09:24.125
That's where to find it.
00:09:24.957 --> 00:09:28.582
If you're already watching or you just found this on audio.
00:09:28.832 --> 00:09:31.457
Hit subscribe or follow wherever you're listening.
00:09:31.582 --> 00:09:32.291
It's free.
00:09:32.457 --> 00:09:36.750
It helps the show and the next episode automatically lands in your feed.
00:09:37.041 --> 00:09:38.957
Okay, let's get back to the workflow.
00:09:39.707 --> 00:09:57.000
Here's the line that actually matters across all five of those steps. There's assisting and there's overreaching Assisting looks like flag draft proposed route to a folder you've already approved.
00:09:58.125 --> 00:10:14.582
Overreaching looks like send without a review delete without confirmation. Commit your calendar with no check-in or make a creative call about which images tell your clients story. My take it's not about how capable the system is.
00:10:15.000 --> 00:10:26.457
It's oversight versus risk. How much oversight does it take off your plate and how much risk does it hand back to you if it gets something wrong.
00:10:27.457 --> 00:10:45.541
A Batch of likely rejects you glance at for 10 seconds, that's low risk a client email that goes out under your name with no review. Completely different risk category doesn't matter if the system nailed it nine times out of ten.
00:10:45.541 --> 00:10:47.750
And one more thing worth saying plainly.
00:10:48.750 --> 00:10:54.207
Does a tool actually replace your judgment as a photographer?
00:10:54.832 --> 00:10:57.750
No, not until someone shows you a source.
00:10:58.207 --> 00:10:59.625
I haven't seen that proof.
00:10:59.625 --> 00:11:02.125
So I'm not dressing it up as fact for you.
00:11:02.791 --> 00:11:12.541
Test the tools that cut friction on the boring steps. Skip the ones that quietly muscle in on the steps that need your eye.
00:11:12.625 --> 00:11:23.416
(Music) So how do you actually decide whether to let one of these into your workflow?
00:11:24.416 --> 00:11:27.707
That's what the framework's for.
00:11:27.707 --> 00:11:35.541
Use it. Test it or Wait, three buckets and here's how I think about each one.
00:11:36.125 --> 00:12:12.875
Use it when the task is low risk repetitive and easy to undo if it goes sideways for example sorting a folder flagging obvious duplicates Drafting a message you'll still read before it sends. These are low stakes easy to walk back and worth the time you get back Test it when the task touches a client, but nothing goes live without you a scheduling proposal waiting on your approval a draft message sitting in your outbox until you hit send.
00:12:14.125 --> 00:12:20.125
This is the category most worth experimenting with supervised reversible.
00:12:20.582 --> 00:12:36.082
You're still the one pulling the trigger Wait is for the stuff you shouldn't hand off period. Final image selection for a client gallery. Anything that touches money without your sign off.
00:12:37.166 --> 00:12:45.207
Anything where the only proof is a vendor telling you it works all of those a definite wait.
00:12:46.375 --> 00:12:51.125
So here's what actually matters when I look at all of this.
00:12:52.166 --> 00:12:55.875
Every task you hand off buys you time.
00:12:55.875 --> 00:13:02.541
But only if you keep a checkpoint at the parts of your workflow where being wrong actually costs you something.
00:13:03.416 --> 00:13:08.375
Which brings me to what I want you to actually try this week.
00:13:09.707 --> 00:13:16.416
This isn't advice from the sidelines, this is what I'm doing with my own workflow this week as well.
00:13:18.082 --> 00:13:24.666
Pick the part of your workflow you dread most. Maybe that's culling a full shoot.
00:13:25.332 --> 00:13:49.332
Grab your next batch run it through one tool that claims agentic culling just one and time it against how long that step normally takes one task one timer one honest before and after and keep the checkpoint in place the whole time. Nothing goes to a client without your eyes on it first no exceptions.
00:13:50.625 --> 00:13:54.332
Did it save you real time without costing you quality?
00:13:54.875 --> 00:13:57.375
Great, you've got your answer build on it.
00:13:58.791 --> 00:14:00.207
Did it not move the needle?
00:14:00.207 --> 00:14:05.041
Worst case scenario you lost 20 minutes nothing you can't spare.
00:14:05.875 --> 00:14:08.125
Now you actually know.
00:14:09.332 --> 00:14:11.332
That's the real ask this week.
00:14:11.416 --> 00:14:15.207
Go try it then come back and tell me how it went.
00:14:16.082 --> 00:14:55.332
And one more thing while I've got you if you want the next episode waiting for you when you're ready on YouTube hit subscribe on your podcast app follow the show Every subscribe and follow tells me you want more of this and that's what keeps me making it. And if a fellow photographer comes to mind who's curious about where AI fits in their workflow. Send this their way Thank you for joining me on Shutter & Algorithms where creativity and technology meet to remind us what it means to be human in a digital world.
00:14:56.000 --> 00:14:59.000
I'll catch you on the next one.
00:14:59.000 --> 00:15:04.250
(Music)