AI and dev agencies

Like a lot of developers, for the last few months all I have been thinking about is AI - and yes - this explains why my posts on Dev to Agency have become fewer in frequency 🙂

I’ve immersed myself in learning all I can about the new frontier of generative AI, and playing with every tool possible. I pivoted my agency consulting role into AI engineer, and started building tech demos of how gen AI can be used to offer new and improved services to clients and enhance their products way beyond what was previously possible.

I’m integrating more advanced and richer AI experiences into my requirements management startup Userdoc, and every week I am blown away by what is possible today.

It’s a freaking exciting time to be a software developer.

Why dev agencies need to pay attention to AI ✨

I ran my software agency when Blockchain was getting big, and when Web3 was going to change the world. But I largely ignored it, and never thought of pitching to my clients how Web3 could help them make money or save money. I felt that when the mass hype died down - it wouldn’t in fact change the world like people thought, and on that point I feel I was correct.

But with AI and large language models (such as GPT-4), it feels different.

AI can literally allow us to do things we NEVER could before. It’s not a new way of doing X, or a faster way of doing Y, it’s an entirely new platform that unlocks completely new opportunities.

Similar to when modern smartphones came out around 2007 - a whole new market segment has just opened up to people who can code and create software.

If 2 years ago a client asked my agency “Hey Chris, we need to chat to our database using natural language, oh and I also want to see graphs and charts!”

I guess I would have known *someone out there could probably do it *- but unless we could find them or a product/API we could leverage, we would have to say “Sorry, this is out of our wheelhouse!”.

But I’ve just built this exact thing in a couple of days, allowing someone to talk to their database (any database) using natural language sentences. Oh, and it supports replying with dynamic graphs and charts - just like magic.

And this is just a drop in the ocean of what you can offer to your clients…

What dev agencies can do today with AI, that they couldn’t do yesterday 💡

OpenAI released API public access to GPT3 way back in November 2021, and I know that specialised AI peeps could weave magic for a long time. But I’m focusing on the generalist developer or agency - and what AI and LLMs can allow us to do a lot simpler than ever before.

The answer to what we can do is essentially endless, but focusing on areas where software development agencies could add value to their clients, here are a few high-level Ideas.

Data mining & semantic search across unstructured datasets - whether it’s ad-hoc Word documents or PDFs, emails, or speech-to-text-recorded telephone calls. AI can add structure and summarisation to this “data”, and also look through it to pull out specific details (e.g. “Which of these recorded telephone calls are complaints about products or services?”
Quick and easy sentiment analysis - this could be used for classifying comments, reviews, questionnaires, forms etc. as Positive, Negative, or Neutral. And even providing more context, e.g. “Samantha G’s telephone call was negative, expressing anger and dissatisfaction about the Tesla she purchased in 2022”
AI Agents - Chatbots have been around for ages (with varying degrees of “AI”) - but whereas chatbots “talk” - agents “do”. Agents perform tasks, and actually get shit done. Agents can actually cancel a customer's subscription, not just tell them how to do it themselves. Agents can actually order you groceries, not just talk about it. Agents are the new generation of chatbots, and agents will be huge.
Automated reporting - managers want a weekly report email about X,Y,Z - often just to see if everything looks ok. But what if AI sent automated report emails when anomalies happened, not using hardcoded boolean logic, but more natural decisions. “Does this week's orders seem similar to the last 10 weeks’ orders”, if not perhaps “Generate a small summary of the differences”, and let’s get a manager to check that out (all automated).
Systems integration - like my database chatting example previously (that generated valid SQL from natural language), LLMs are amazing at understanding code and syntax. You can give them an API spec and they can generate example integration code. You can give them code, and they can generate a detailed spec. This opens a lot of opportunities for writing dynamic integrations and allowing the LLM to be the middleman.
Content generation - I’m not talking about blog posts or product taglines. But boring business content that needs producing. Forms, reports, documents, and even emails. Things businesses spend time on that now could be automated.

So what should I do about it? 🤔

I believe development agencies are in a great position to learn, master, and offer AI services.

Even though it seems we are at “peak AI excitement” at the moment, we are not.

As technologists, we see it coming first. Give it a year, maybe two years - and every single business we will want to harness the power of AI for two core reasons that are as old as time.

To save money
To make money

Sure, there will be AI-assisted products and SaaS services to help them, but just like always, organisations have endless business rules and complexity, and many will still need custom software to work exactly the way they need it to.

How do I get started? 🏃‍♂️

Start learning by playing with AI development tools, AI is moving crazy fast, and every day there are new tools, models, and companies that provide opportunities and inspiration.

I don’t just mean ChatGPT, I mean tools like LangChain and BabyAGI that let you talk to many different models in many different ways. This is what opens open the power to developers.

Build small demos, build them for fun, build them to explore what’s possible. It will inspire other ideas and help you get a grasp on the true power available.

For example, recently I got beta access Anthropic's Claude API, which allows you to send 100k worth of tokens in your prompt. I didn’t get the power of this at first, but I now see it. And it’s huge. Being able to send around 75,000 words in your LLM prompt is a game changer, and it’s filled my head with new ideas.

Find a way that works for you to stay up to date. I listen to AI Podcasts and read AI newsletters - and follow people in the AI space on Twitter such as swyx.

Have fun, exciting times ✨