Some days, working in AI feels like moving across quicksand.
I find solid ground, get comfortable, and then something changes. A new feature appears. A tool that couldn’t handle the job last month suddenly can. I have to find my footing again.
I find that exciting. It also means part of my day is spent evaluating the tools, not simply using them.
When I say I’m into AI, I mean I have two computers in front of me. On the Mac, I have ChatGPT Pro. On the PC, I also have ChatGPT Pro. I work on several projects and conversations at the same time across both computers. I also use Claude Code.
I barely type anymore. I talk to the AI, and I use tools such as Wispr Flow to put my words into the software I’m working in.
The distinction for me is where those words go. Instead of simply recording something to transcribe later, I’m speaking into the task in front of me. Flow turns my speech into written input. In an AI chat, that becomes my request for the AI to work on. I still need to read what came through and check the result. I explain that workflow in Why AI adoption fails.
I’ve spent thousands of hours exploring AI for our clients, for our own work and because I’m interested in what it can do.
Some of that work involves building software. Some involves websites, videos, spreadsheets and presentations. Some is about finding out whether the marketing we’re doing is actually bringing in business.
Rather than give you a list of AI tools, I’ll divide it into three areas: marketing, administration, and sales and leads. These are the same areas we work across at Salient AI.
You probably don’t need my desk setup. But you might recognise a few problems we’re working on.
1. Marketing: creating the work and measuring whether it helps
Tracking Google rankings and visibility in AI answers
One of the applications we’re building is a rank tracker.
It tracks keywords, search volume and rankings. We’re also tracking mentions in ChatGPT and Google’s AI Mode, along with citations that point back to websites.
We’re doing this because we want to keep analysing our marketing.
If we improve a website or publish content, is the business becoming easier to find? Are the changes helping? Where do we need to put more effort?
Traditional rankings tell us part of that story. Mentions and citations in AI answers give us another part to examine.
There are limits. One answer doesn’t represent every answer someone might receive. We’re looking for useful patterns over time, rather than treating one mention as proof that the job is done.
Building websites and producing social videos
We’re also designing websites using ChatGPT and Claude Code.
The business still needs a clear message, useful information and an obvious next step. AI helps with the building, but somebody has to decide what belongs on the website and whether it makes sense to a customer.
Social media is another part of my day.
We’re using computer-use tools and open-source programs on the computer to create videos. We’re also connecting that work with automated posting workflows for clients.
That brings together the idea, footage, sound, text, finished video and accompanying post.
The question is whether the result will interest the right person. A video can be finished and still have a weak opening or an unclear message.
Think of Mad Men. The creative idea gets your attention. In our work, we also need to follow what happens after someone sees it: do they visit, call or become a customer?
These projects sit within our AI marketing work: helping people find the business, understand it and take the next step.
Re-evaluating the tools I pay for
I recently stopped my Adobe subscription. I’ve started using a free suite for graphics and vector work, and I’m looking at what other freely available tools can do.
That was a decision about my own work. It doesn’t mean every business should cancel the software it relies on.
But it does mean I’m willing to ask whether the tools I used before are still the right fit. There are people around the world building useful software and making it available for free. Evaluating those options has become part of the job too.
2. Administration: removing the small jobs that keep getting in the way
A phone app that puts job photos where they belong
Here’s a simple example.
One client has many staff who take pictures and videos. Those files need to be connected to the right projects and made available to the company.
Sending them around by email adds work.
It has a bit of that Office Space feeling. The TPS report is familiar, but here it’s a photo someone has to send, find and file. The little steps keep adding up.
So we created a small phone-based tool connected to Google Drive. Employees open it, take their pictures or videos, and the files are allocated to the right projects and pushed to Drive.
The aim is simple: capture the information once and get it where it belongs.
This uses a PWA, short for progressive web app. It’s built with web technology but can look and feel like an app on your phone. You can make it available without putting it in Apple’s App Store or Google Play. Phone features and installation options vary by device and browser. MDN’s PWA guide explains the approach.
For this particular tool, the build took less than an hour. Maybe 30 minutes.
That’s one example, not a promise that every business application takes half an hour. But it shows why I spend time looking at these small points of friction. A fairly simple tool can change how information moves through a company.
Exploring financial software
We’re also building financial software.
Part of the ambition is to see whether we can replace work we currently pay a bookkeeper and accountant to do.
That’s a project we’re exploring, not a claim that we’ve already replaced every part of those roles.
We’re attaching agents and bots to financial-analysis work and looking at how the information can be brought together and checked. The useful result would be less time assembling records and a clearer view of what is happening in the business.
The output still has to make sense. If the records are incomplete or a calculation is wrong, a polished report won’t solve that.
Working with the files and systems already in the business
There’s also the everyday work: modifying PowerPoint presentations, working on Excel sheets and building applications for ourselves and clients.
I do work with miRoncol, where we’re building parts of the customer journey. That involves coding, deploying on Amazon Web Services and using GitHub for version control, security workflows and deployment.
Version control keeps track of software changes. Deployment puts a version where it can run. These are parts of managing an application beyond the first working screen.
GitHub supports automated build, test and deployment workflows, but the project still needs the right setup and review. GitHub’s documentation explains those capabilities.
Across all of this, the administrative question is similar: where does work get held up, repeated or passed around unnecessarily?
That’s the starting point for our AI administration services.
3. Sales and leads: connecting the inquiry to the business result
Bringing two systems into one CRM
Another project is a new CRM system that consolidates two pieces of software.
A CRM is where a business keeps track of customers, inquiries and the work needed to move those relationships forward.
Think of the familiar feeling of walking into Cheers. People know who you are and remember the conversation. A useful CRM helps the team bring that kind of context to the next customer interaction, using the information actually recorded.
When information is split between systems, people have to bridge the gaps. Combining the right parts can make the process easier to follow.
The goal is a more useful way to manage the work, rather than another place to enter the same information.
Working on a new point-of-sale system
We’re also creating or modifying a point-of-sale system for a client.
That’s the software used to handle a sale. It’s another example of using AI-assisted development to work on a specific business need.
A POS, a CRM and a website all touch different parts of the customer experience. Looking at them together helps us see where the process becomes awkward.
Connecting phone calls with website traffic
We’re analysing phone calls for clients and bringing that information together with website traffic.
We want a more complete view of how marketing investment translates into inquiries and conversions.
More website visitors can look encouraging. But what happens next?
Do they call? What are they asking about? Are those inquiries relevant? Where does the process lose people?
The available information won’t explain every sale. It can still help us move beyond traffic totals and ask better questions about business results.
That’s the connection between marketing and our sales and lead-support work.
The tools keep changing. So does the work.
Alongside those projects, I spend time experimenting.
I’m exploring Hermes agent on a rented server and locally on a computer, looking at how it could coordinate tasks with my ChatGPT setup and create assets through repeatable workflows.
I’m also trying to manage more from my phone. I’ve connected the setup to Discord so different project channels can keep the work separated.
Those are experiments. I’m still exploring what is useful and reliable enough to keep.
The same ongoing evaluation applies to ChatGPT, Claude Code and Gemini. If something doesn’t work for me today, I don’t assume that answer will hold forever. I may need to try it again when the capabilities change.
That’s also why I’m cautious about committing to a year of software just because it looks useful today. My preference is to weigh the annual savings against how likely my needs and the available tools are to change.
We’re looking at how our staff work with AI too. I think people who have begun trying it will be better placed to understand how their work is changing.
I was talking to a colleague, Carl, who put it in a way that stuck with me. Six months ago, he didn’t know how to program. He didn’t know what GitHub was. Today, he’s using programming tools for his business.
That took more than access to an AI tool. He has been practising every day, learning the tools, learning the framework and working out how the pieces fit together. That takes time.
When people tell me they use AI every day, they may mean chatting with it, asking questions or getting help with a few simple tasks. Those are useful starting points. But there’s more to explore.
My concern is that if we never go beyond those first uses, we could miss how much the work around us is changing.
Take programming. I think more of the job is becoming planning, connecting the pieces and checking the result. Knowing how to write code and knowing how to design a whole system are different skills. That makes learning and judgment more important, not less.
That’s what keeps me interested. Even after thousands of hours, there’s more to explore, and the ground keeps moving.
Practical AI coaching for your business or career
This is also where I can help as a practical AI coach.
I prefer to learn by watching someone do a task and then getting hands-on myself. That’s the approach I want to bring to coaching. We work through a relevant project, from explaining what we need to checking the finished piece. You see the corrections and decisions in between, then practise doing it yourself.
For a business owner, we can start with a recurring frustration: photos arriving by email, information split between two systems, or marketing reports that don’t explain where the leads went.
For someone looking for work or adapting to a changing role, we can start with a task that matters in the kind of job they want. Practise it, learn how to check the result and build an example they can explain to an employer.
The aim is to develop useful skills and confidence through doing. Nobody can promise that learning an AI tool will get you hired, but being able to show how you use it gives you something concrete to talk about.
You don’t need to copy my desk. Start with something relevant to your business or career, and let’s work through it together.
Talk through a practical AI starting point.
Created with AI assistance.