Thomas McCarthy-Howe, CTO Vconic, vCon Father

Rob Pickering, Internet and Real Time Communication Software Expert,

RJ Burnham, Founder and CEO Consig AI

João Camarate, Entrepreneur, Investor, Startup Nerd, Advisor to Broadvoice

Lyle Pratt, Founder CEO Vida Voice

Introduction

I follow the above thought leader’s Linkedin posts and our one-on-one conversations about AI and the role it plays in software and product development.

In a world choking on AI BS, their posts and our conversations help lower my blood pressure.

The above leaders are engineers, their thinking is shaped by decades in engineering. They are guided by truths: complexity should be avoided where possible. Their experience in AI matters, and is worth all our time to listen.

Rob’s recent comment on how AI changes the landscape for developers young and old was great. I receive lots of questions about this from pre and post graduate developers, young and old. They are worried about their futures, many struggle to get hired.

Joao has been doing impressive work on AI-enhanced development. I consider his work essential knowledge for all technologists to own their future.

This panel session will bring together the panelists’ experiences and what they see as the important changes in coding and product development; and we’ll try to keep it as real as possible with some recommended practical steps for developers old and young.

Time flew on this podcast, there were so many sagely insights. I’ll try and highlight the ones that stuck with me. I apologise that its 47 minutes long. It’s worth every minute of your time to listen.

Each of the panelists had the same questions: what are you working on, and what are your preferred software / product development tools.

RJ Burnham, Cursor

RJ kicked off, Consig is focused on Voice AI in Healthcare. I recently discovered I have Lyme disease for about one decade. Fatigue was not one of my symptoms. I have lots of recent firsthand experience on the challenges the medical profession face in patient communications given the pollution of the PSTN, Consig is addressing a market need, and its timing is perfect.

One of RJ’s pet peeves is how AI is often used for the benefit of the provider, not the patient. Here he thinks Voice AI can be used to benefit both parties.

On tool sets, it depends, Cursor is the workhorse. For all people aspiring to become developers, learn to use Cursor. Senior developers are using it today.

Around the tool are customer / user feedback organized into the specs on what will be written. RJ highlights Product Engineering is well served by Cursor. Whilst Project Management remains a gap. BTW, we have a TADSummit online Conference session in November with Jason Goecke on Spec Driven Development.

RJ gives a great quote, “the art of writing code is no longer a critical skill.”

Lyle Pratt, Codex

Lyle makes an important point on the rate of change. “Even if all LLM progress were to stop, the entire enterprise software market will change in the coming decade.” The cat is our of the bag on AI software development, and that trajectory will accelerate.

Vida develops AI solutions for SaaS businesses like smartmoving.com. And their customers are pushing them for AI capabilities. The end customers are driving demand for Vida’s solutions.

The addons could be billing integration, or using SIP registration for AI agent registration. Vida are integrating their software into existing SaaS partners. Enabling a roofer’s AI agent to answer calls, business queries, and even bid on work.

“The AI Agent needs to do real work in the software you’re using today.” Lyle Pratt.

Vida uses OpenAI’s Codex. OpenAI claims that Codex can create code in Go, JavaScript, Perl, PHP, Ruby, Shell, Swift, and TypeScript, though it is most effective in Python.

Senior devs have benefited most from the shift to AI-enhanced software development, as they bring experience and the tooling. But as we’ll discuss later, junior devs should take heart on the shifts over the coming years that will benefit them.

Lyle makes a good point often repeated, we’re going to see billion dollar companies with teams. He explained this through an analogy of less actors, but millions of directors.

João Camarate, Claude Code

João has used his time since selling his last company to build AI expertise across the product development lifecycle. Initially focused on building a new product, but recently has focused on helping his portfolio companies, friends, and larger companies with for example 1500 developers transition to AI-enhanced development.

The preferred tools depend on the day, it’s a rapidly developing market, the tooling gets better, and the skills of the users also improves. Senior developer productivity improves with leaps and bounds over the passed 18 months. AI development is no longer optional.

For those who follow Joao, his go to stack is not a surprize, Claude Code (generally backend) and Lovable (front end, ideation, rapid prototyping). Though tools are interchangeable. There is no one correct answer. Many of the available tools create impressive code.

Joao has been examining how the processes and culture changes with AI-enhanced development. He gave an example of a start-up that through the use of AI has no backlog. They do not want to clutter their product with features for feature’s sake. Code has become “costless”.

I asked how the AI tools applied to an organization with 1500 developers. I get the benefit for a small team punching above their weight class. But for a large team, I’ve not seen success cases.

Accelerating development hits a limit within large organizations. The conversation moves onto removal of functions, do we need product owners and product managers? The few examples I’ve seen in larger organizations resulted in a downsizing in the development team.

Rob Pickering, Cursor

Rob uses the term internet plumber, but he’s so much more than that! He also uses the term AI media streams, rather than voice AI. His favorite tool is Cursor. I fully understand Rob’s path from VSCode to Cursor. It reflects tools we are accustomed to. With Claude Code as a number 2.

Rob makes a funny point, AI development is the worse it will ever be, it’s only going to get better. Today getting the tool to plan and chunk the plan into executable pieces is the recipe for success. And that’s only going to get easier.

For large projects the pull request automation tools used in open source projects are good, and directly applicable. Rob thinks AI coding tools enable the refactoring of technical debt, as commodity tasks have been done many times. He sees this as a theme in the coming years, as large code bases are refactored. Exciting times!

Rob discusses his experiences in building code himself versus AI. How he uses a REACT framework for UIs, and has the AI tool update the framework, while he’s working on an API update. Spending time on the work he enjoys and adds most value. This is a great share on the reality of AI coding in practice.

Rob highlights a shift in developer training. Junior devs are going straight into using the AI coding tools. Skipping the Computer Science degree, internships, and development paths. We’ll discuss later what this means in skills young developers should foster.

AI Coding is getting better at re-architecting big projects, so good that it will be BAU (Business As Usual) soon. The wall senior developers have will slowly lower as their specialized knowledge ages out of the code base. Rob posits we may soon see AI tools rewrite / refactor / simplify REACT, and make the experience much better, and built for a better fit for AI coding tools.

Thomas McCarthy-Howe, Cursor

Thomas’s work focuses on vCon, mostly plumbing like proxies and conservers. He was working during this podcast on an MCP server using Cursor (Model Context Protocol, acting as an intermediary to connect AI models with tools and data). Cursor is generally front end with Claude for backend coding.

He plans to never write another unit test or Git commit in his life!

A shift Thomas sees is who he writes documentation for, its for robots to use his libraries. His team standardized on Supabase, a postgres development platform. To the tools mentioned earlier, Lovable, an AI-powered application builder, works with Supabase, work hand in hand.

Documentation is written inline with the code, for the code. Always an aspiration in coding, but now done for real with AI coding.

Thomas raised many questions for AI coding. Is it OK to generate the documentation automatically (yes), when is appropriate to use AI coding? (everywhere)

General Discussion on AI Coding

Seeing high schoolers raised on AI tools, AI coding is approached with the attitude, “why would you do it anyother way?” Even though they do have Scratch and Python experiences, they are distant memories.

Joao raised his experiences with security companies. They are under immense threats from AI enhanced hacking. And this leads to accountability. Do we expect a human to be ultimately accountable for testing software. I know some people, when it’s their name on the code will spend the time and do the checks. But I know some high schoolers that assume the AI checks are good enough. And in some respects better than a human, or they can do.

RJ highlighted the importance of people who understand what your customers want and need; a technical product manager that extracts mines of information from customers with effective questioning. AND someone who has the wisdom (experience) to use those requirements to guide the technical architecture.

RJ frames this as once the context window is large enough, the whole project including experience based architecture is within the grasp of AI coding. I think this is going to be a stretch, but given the improvement in the tools over the last 18 month, it’s not unreasonable expectation.

Lyle framed the problem as, asking normal people to prompt engineer. It’s not possible, that’s the service Vida provides to its customers, so the agent is reliable and reproducible. The AI agent does the work required by the end customer. Now we can train people to do a task, its managing intelligence, so the end goal is achievable. But we’re not there yet, but the vision is clear.

I wrapped up asking about one piece of advice you’d give a junior dev trying to break into the software industry. Rob kicked of with grab AI coding tools and build skills by playing with them. They are great learning tools. When you’re doing background projects, make sure the tools are AI based so you can play with them,

RJ highlighted the importance of nurturing junior devs, they need support through mentorship. Joao explained the popular stereotype of a developer as a hoodie, headphones, fed pizza, and given a spec to code. Today, almost the opposite is required, it’s a love of building products that solve customers’ problems, as there are now existing tools for creating the code.

Thank you Thomas, Rob, Lyle, João, and RJ for a fun, insightful, and uplifting discussion.