Ideas Taste and Execution
Throughout my career as an architect and engineer of various systems over the last 30 years I've had a ton of ideas. My consulting company Mystic which has been in full effect for 25 years now was the hatching of an idea to bring along the best people in various disciplines I've run into and deploy them with me to help bring solutions to companies around the world. We've been able to solve problems for a lot of companies over the years such as Twitter 1.0, Motorola and Airbus to name a few. In each of these instances we've found our ultimate success by taking the seed of an idea on how to solve it and taking it the steps required to execute successfully.
If you read enough books on entrepreneurship and business, or listen to interviews with successful founders it becomes evident that what sets apart a successful endeavor is great execution, and a bit of luck. At no time was the idea the most important factor of the success.
Instagram was a location-based check-in app like Foursquare -- remember that app?
Slack was an internal comms tool for a gaming company (oddly enough Flickr was also an online game that pivoted to photo sharing)
YouTube was a video dating site where people could upload videos describing their ideal date
These 3 examples should be enough proof that what mattered here was execution.
Enter AI and "agentic" coding.
I have seen a big trend online where "founders" or perhaps "marketers" are trying to convince you of something.
Ideas are now gold!
Taste is what matters in this new economy!
Whether it's a VC talking about engineers who code with AI and write 10k lines of code a day, or a course seller on "how to effectively prompt engineer your way to a million dollar app!". When it sounds like someone is trying to sell you a scheme, or convince you of something to further their own agenda, it's probably just that.
Since the beginning of the internet boom in the 90s the cost of turning your ideas into a reality meant hiring at a minimum designers and engineers to get a prototype together. The friction turned many away whether due to not having capital or not being able to convince a VC to fund it. With AI the friction according to those that are peddling that "ideas and taste" are all you need is effectively gone! The truth is a bit more nuanced than that.
At various times throughout my career I've been hearing how no-code solutions were going to kill software development for good and enable business owners to turn their ideas into reality. Whether it was Visual Basic or HyperCard in the desktop era, Dreamweaver or Flash in Web 1.0 or something as low level as BizTalk which promised to connect disparate systems together with ease, the reality was often barely functional or underlying it all extremely complex.
It might seem to some like the solutions out there with AI means we may finally be there. I know for me, the first time I used Claude Code to create an admin dashboard for a project, I was awestruck. It looked amazing. It was akin to hiring a few junior frontend engineers and having them build something that looked great after a few minutes of conversation.
And then I looked at the code it wrote.
The code was being held together by duct tape and chicken wire. What followed was countless hours of prompts and tinkering with the context to approach something passable. It didn't require vibes but much more expressive requirements, more guard rails and a set of instructions on acceptable code quality and standards. It often feels like it's performing magic but ends up more like an apprentice who sometimes turns your code into a goat. One of the most beautiful things to emerge from this process was that I have tightened up how I explain things in requirements and I ask better questions.
Building an app with an LLM is like someone who can't even boil water watching a cooking show demonstration and thinking they can run a restaurant. The real work lies in the edge cases and the day to day: suppliers, staff and health inspectors. It's no different with software. For this dashboard, It took engineering effort and a lot of time to turn the prototype that was built and bring together the knowledge and execution necessary to build a great application.
Large language models in software development is another tool in a software engineers toolbox that can be used toward creating a great product. It can be a great way to leverage the giant body of knowledge that it has to your own ends. It doesn't write perfect code, and it sometimes doesn't even produce working code! It will sometimes delete your code when it encounters an error with a new feature (git commit is your friend). It will ask you for every new session if it's allowed to do something until you learn how to configure the guard rails.
It is the junior engineer who seemingly knows everything and is confident they do, and often times gets it apologetically very wrong. It took me thirty years of experience to tell the difference.
So how can you leverage AI in your organization today?
1. Work together with your team to set up initial context and memory for your individual project and your organization.
2. Code review is now where a lot of care is needed. Your engineers are going to use AI to write code, care should be taken to make sure they know what it did and if it is fulfilling the requirements and user stories effectively.
3. Nurture the junior developers on the team so they can level up their skills and recognize good clean code when they see it.
4. Enable your business folks and product managers to build out prototypes to their hearts desires, and then bring these to engineers to find out what feasibility actually looks like.
The future has always been so very fun and interesting. Those that are telling you that AI will democratize execution are stuck in a hype cycle that hasn't explored why good execution is the key differentiator between a neat looking tasteful app and a thriving business.
If you'd like to engage with the team and find out how we can help you with your next project. Reach out to us today.