WEBVTT
00:00:00.239 --> 00:00:03.919
Welcome to the Inspiring Tech Leaders podcast with me, Dave Roberts.
00:00:04.160 --> 00:00:09.519
Today I'm going to look at vibe tooling and compare GitHub Copilot, Replic, Cursor, and Lovable.
00:00:09.919 --> 00:00:12.080
The idea of vibe tooling is simple.
00:00:12.320 --> 00:00:20.719
Instead of sitting down and writing thousands of lines of code yourself, you describe what you want to build and an AI system works out how to build it.
00:00:20.960 --> 00:00:29.440
You can write the code, create the files, connect databases, run tests, identify errors, make changes and increasingly deploy the finished application.
00:00:29.679 --> 00:00:33.359
And there are now several serious players competing for this space.
00:00:33.600 --> 00:00:39.359
They all use AI to help us build software, but they approach the problem from very different directions.
00:00:39.600 --> 00:00:46.640
And that distinction is becoming increasingly important because the question is no longer simply which AI coding assistant is best.
00:00:46.880 --> 00:00:54.159
The much more interesting question is what kind of software builder are you, and what sort of relationship do you want to have with the AI?
00:00:54.399 --> 00:00:56.560
Let's start with the term vibe coding.
00:00:56.719 --> 00:01:01.119
The phrase became popular because it captured a very different way of creating software.
00:01:01.280 --> 00:01:10.560
You have an idea, you describe it in natural language, the AI generates the code, you look at what has been produced, tell it what you want changed, and continue iterating.
00:01:10.799 --> 00:01:23.519
You might say, Build me a custom portal, add authentication, give it a dashboard, connect it to a database, add a search function, make the interface look modern, and add in an administration area.
00:01:23.760 --> 00:01:32.640
Historically that would have been a significant software development project, but with today's tools it's becoming a conversation, and that is the important change.
00:01:32.879 --> 00:01:36.400
We are moving from writing software to directing software.
00:01:36.640 --> 00:01:45.280
The human increasingly becomes the person who defines the problem, describes the desired outcome, makes decisions and judges whether the result is good enough.
00:01:45.519 --> 00:01:51.040
The AI is increasingly becoming the implementation engine, but there is a big catch.
00:01:51.280 --> 00:02:01.120
The various different platforms don't give you the same experience, and I think the easiest way to understand them is to imagine them as four different people building the same application.
00:02:01.599 --> 00:02:04.480
Imagine you want to build a simple business application.
00:02:04.640 --> 00:02:13.439
It has a web interface, users need to log in, there's a database, there is an administration dashboard, and there are a few integrations with external services.
00:02:13.680 --> 00:02:19.360
If I give that requirement to Lovable, I'm essentially saying here is the product I want, build it.
00:02:19.680 --> 00:02:27.439
Lovable is particularly focused on taking an idea expressed in natural language and turning it into a working full stack application.
00:02:27.680 --> 00:02:39.680
Its current approach can handle the front-end, back-end authentication, database and integrations, and its agent can plan more complicated tasks, work through multiple steps, and test the application.
00:02:39.919 --> 00:02:45.599
That makes Lovable perhaps the clearest expression of what many people mean when they talk about vibe coding.
00:02:45.759 --> 00:02:51.039
You don't necessarily start by thinking about the technology stack, you start with the product.
00:02:51.280 --> 00:02:54.240
You might say, I want a dashboard for a sales team.
00:02:54.479 --> 00:03:00.400
It needs a login screen, a customer list, sales chart, and an area where managers can see performance.
00:03:00.560 --> 00:03:04.159
And Lovable starts turning that description into an application.
00:03:04.400 --> 00:03:06.159
The appeal is obvious.
00:03:06.400 --> 00:03:18.400
If you're a founder, product manager, designer, business analyst, or technically curious person who doesn't necessarily want to spend hours working through configuration files, frameworks, and dependencies, this is incredibly powerful.
00:03:18.639 --> 00:03:24.960
You can concentrate on what you want the application to do rather than exactly how the applications should be constructed.
00:03:25.120 --> 00:03:28.719
And Lovable has pushed this further during 2026.
00:03:28.879 --> 00:03:37.599
Its autonomous capabilities allow it to take larger pieces of work, plan them, and continue working without requiring you to micromanage every step.
00:03:37.919 --> 00:03:40.479
There is another interesting development here as well.
00:03:40.800 --> 00:03:48.960
Lovable has introduced ways for applications created on the platform to work with AI tools such as ChatGPT and Claude through MCP.
00:03:49.120 --> 00:03:58.319
So we're beginning to see a world where AI simply isn't helping us build applications, AI can also become a way for users to interact with those applications.
00:03:58.560 --> 00:04:00.960
That is quite a significant shift.
00:04:01.280 --> 00:04:03.120
Now let's move to Replit.
00:04:03.360 --> 00:04:05.919
Replit takes a slightly different approach.
00:04:06.240 --> 00:04:16.399
It is still very accessible and very focused on describing what you want and allowing an agent to build it, but it gives you more visibility into the underlying development environment.
00:04:16.720 --> 00:04:25.199
Replit's current agent can build applications from natural language and the platform combines development, execution and deployment in one environment.
00:04:25.439 --> 00:04:37.439
Its Agent 4 release in 2026 emphasises autonomous work, parallel agents, and the ability to deal with things such as authentication, databases, back-end functionality and front-end design.
00:04:37.680 --> 00:04:46.000
So if Lovable feels like you're saying Build me an application, Replit feels more like saying, Build me an application, but let me look under the bonnet.
00:04:46.240 --> 00:04:47.680
That distinction matters.
00:04:48.000 --> 00:04:52.800
Replic can be an excellent bridge between the traditional developer and the non-developer.
00:04:53.360 --> 00:05:00.000
You can start with a natural language prompt, get an application running very quickly, and then start exploring the code.
00:05:00.240 --> 00:05:02.639
You can see what the AI has actually created.
00:05:02.800 --> 00:05:08.720
You can experiment, you can run it and change it as needed, and perhaps most importantly, you can learn from it.
00:05:08.879 --> 00:05:17.199
That makes Repli particularly interesting for people who want the speed of vibe coding without completely abandoning their traditional software development experience.
00:05:17.439 --> 00:05:19.920
Imagine someone who has an idea for an application.
00:05:20.079 --> 00:05:28.160
They may not know Python or JavaScript, they may not understand databases, they may not know what an API is, but they can describe their idea.
00:05:28.399 --> 00:05:32.639
The AI creates something, they see it working, they start asking questions.
00:05:32.879 --> 00:05:34.240
What does this file do?
00:05:34.399 --> 00:05:36.240
Why have you created this database?
00:05:36.480 --> 00:05:38.399
What happens if I change this?
00:05:38.720 --> 00:05:42.160
That is a completely different way of learning software development.
00:05:42.399 --> 00:05:50.240
Instead of learning syntax first and building something later, you can build something first and then learn the underlying technology as you go.
00:05:50.480 --> 00:05:55.279
And that's why I think Replit sits somewhere between vibe coding and traditional development.
00:06:34.139 --> 00:06:35.740
Now we come to Cursor.
00:06:35.980 --> 00:06:41.819
Cursor is probably the tool in this comparison that is closest to the traditional professional software developer.
00:06:42.139 --> 00:06:52.780
Cursor is an AI-powered development environment built around the idea that AI should understand your existing code base and become an agent capable of making substantial changes across it.
00:06:53.100 --> 00:07:05.020
Cursor describes itself as a coding agent for building ambitious software, and its current platform supports autonomous agents, code-based understanding, multiple models, cloud agents, and parallel development.
00:07:05.259 --> 00:07:07.500
And this is an important distinction.
00:07:07.819 --> 00:07:16.620
Cursor doesn't primarily say give me an idea and I'll create your application from scratch, it says give me your software and let me help you develop it.
00:07:16.860 --> 00:07:24.220
Of course, Cursor can absolutely build new applications, but its real strength becomes apparent when you already have a serious code base.
00:07:24.780 --> 00:07:42.540
You might have millions of lines of code, you might have a complex React application, you might have a Python backend, you might have hundreds of tests, you might also have authentication, APIs, databases, and deployment pipelines, and you can ask Cursor to understand that environment and then make a significant change.
00:07:42.780 --> 00:07:55.579
For example, you could say add multi-factor authentication to the application, update the login flow, modify the database schema, update the tests and make sure the existing authentication behaviour isn't broken.
00:07:55.980 --> 00:08:00.060
This is a very different proposition from simply generating a landing page.
00:08:00.300 --> 00:08:03.980
Cursor has already been pushing heavily into autonomous agents.
00:08:04.220 --> 00:08:15.579
Its cloud agents can operate in isolated environments, make changes, test the software, and produce artifacts such as screenshots, videos and logs so that a developer can review what happened.
00:08:15.819 --> 00:08:18.939
And this is where the term vibe tooling becomes useful.
00:08:19.180 --> 00:08:25.100
Because the human is no longer necessarily sitting there watching the AI type, you can give an agent a task.
00:08:25.259 --> 00:08:29.020
It can go away and work on that task and then come back with the results.
00:08:29.180 --> 00:08:31.980
You review the changes and decide whether to accept them.
00:08:32.299 --> 00:08:39.740
Cursor has been moving towards fleets of agents working in parallel, where different agents can work on different pieces of the code base.
00:08:39.980 --> 00:08:46.139
That begins to look less like an AI autocomplete tool and more like an AI software engineering team.
00:08:46.379 --> 00:08:48.860
Now let's talk about GitHub Copilot.
00:08:49.259 --> 00:08:56.939
This is where the comparison becomes particularly interesting because GitHub Copilot has evolved enormously from the product many people originally knew.
00:08:57.100 --> 00:09:01.419
The old model of GitHub Copilot was essentially an AI pair programmer.
00:09:01.579 --> 00:09:05.179
You start typing the code and it suggests the next piece of code.
00:09:05.340 --> 00:09:06.939
You accept it and continue.
00:09:07.179 --> 00:09:11.259
That is still part of GitHub Copilot, but it's no longer the whole story.
00:09:11.419 --> 00:09:21.659
GitHub now has copilot agent mode inside development environments where the agent can determine which files need to be changed, execute commands, iterate and remediate problems.
00:09:21.819 --> 00:09:24.379
And then there is GitHub Copilot Cloud Agent.
00:09:24.699 --> 00:09:27.979
This is where GitHub's strategy becomes particularly powerful.
00:09:28.219 --> 00:09:33.500
You can give GitHub Copilot a task and let it work in the background in its own development environment.
00:09:33.659 --> 00:09:41.659
It can investigate the repository, make changes, run tests, and ultimately produce a pull request or leave the work on a branch for you to review.
00:09:41.979 --> 00:09:47.019
So imagine you have a GitHub repository with an issue saying improve the search experience.
00:09:47.259 --> 00:09:53.979
Previously, a developer would read this issue, investigate the code, make changes, run tests, and create the pull request.
00:09:54.139 --> 00:09:57.339
Now you can increasingly assign that task to an agent.
00:09:57.579 --> 00:10:01.099
The agent investigates, plans, writes code and tests.
00:10:01.259 --> 00:10:03.019
The human then reviews.
00:10:03.339 --> 00:10:06.859
That is a fundamental change in the software development workflow.
00:10:07.099 --> 00:10:16.939
GitHub has also added things such as model selection, self-review, security scanning, custom agents, and CLI handoff to its coding agent experience.
00:10:17.259 --> 00:10:21.899
And this brings us to what I think is the most important difference between the four tools.
00:10:22.219 --> 00:10:25.979
Lovable is extremely focused on turning an idea into an application.
00:10:26.219 --> 00:10:31.500
Repli is focused on making that process accessible while keeping the development environment visible.
00:10:31.740 --> 00:10:35.659
Cursor is focused on making an existing code base highly agentic.
00:10:35.819 --> 00:10:43.819
GitHub Copilot is increasingly focused on putting AI agents directly into the software development's lifecycle and GitHub workflow.
00:10:44.059 --> 00:10:45.579
So which one is best?
00:10:45.819 --> 00:10:47.979
Well, the answer is it depends.
00:10:48.219 --> 00:10:55.019
For example, if I'm a non-technical founder and I have an idea for a SaaS product, I would probably start by looking at lovable.
00:10:55.339 --> 00:10:59.099
I don't necessarily want to understand the architecture on day one.
00:10:59.339 --> 00:11:07.259
I want to see the product experiment and then have the ability to say, that's not quite right, change this, add that, or remove this.
00:11:07.500 --> 00:11:10.139
That's where lovable becomes extremely compelling.
00:11:10.379 --> 00:11:15.899
If I want more visibility into the code and the development environments, Replic becomes very attractive.
00:11:16.059 --> 00:11:17.819
It's a great middle ground.
00:11:18.139 --> 00:11:22.859
It allows you to move more quickly while still giving you access to the underlying machinery.
00:11:23.099 --> 00:11:29.019
If I'm a professional developer working on an existing software product, Cursor becomes extremely compelling.
00:11:29.259 --> 00:11:37.740
The ability to understand a code base, work across multiple files, use different models, and delegate substantial engineering tasks is very powerful.
00:11:37.979 --> 00:11:49.500
And if my organisation is already heavily invested in GitHub, GitHub actions, pull requests, issues, security controls, and enterprise deployment processes, GitHub Copilots has a huge advantage.
00:11:49.740 --> 00:12:00.059
It isn't just an AI coding tool, it sits in the software development ecosystem, and that matters, because enterprise software development isn't simply about generating code.
00:12:00.219 --> 00:12:07.259
It's about governance, security, identity, source control, code review, testing, auditability, and deployment.
00:12:07.500 --> 00:12:12.859
It's about understanding who changed what, why they changed it and whether that change is safe.
00:12:13.099 --> 00:12:18.699
This is one of the reasons I think the enterprise battle around AI coding is going to be particularly fascinating.
00:12:18.939 --> 00:12:22.859
The winner won't necessarily be the tool that writes the best piece of code.
00:12:23.099 --> 00:12:26.859
It could be the platform that creates the best overall development system.
00:12:27.099 --> 00:12:30.139
And that brings us to one of the dangers of vibe coding.
00:12:30.379 --> 00:12:34.939
Just because an AI can build something doesn't mean the thing is built is good.
00:12:35.179 --> 00:12:38.539
This is probably the biggest misconception surrounding the whole movement.
00:12:38.779 --> 00:12:41.019
You can ask an AI to build an application.
00:12:41.179 --> 00:12:45.979
It might work, it might look fantastic, it might even have thousands of lines of code behind it.
00:12:46.139 --> 00:12:48.139
But does it have a sensible architecture?
00:12:48.299 --> 00:12:49.259
Is it secure?
00:12:49.419 --> 00:12:51.099
Does it handle edge cases?
00:12:51.259 --> 00:12:53.339
Are permissions correctly implemented?
00:12:53.579 --> 00:12:55.659
Is the database designed properly?
00:12:55.899 --> 00:12:56.859
Can it scale?
00:12:57.099 --> 00:12:59.019
Does it leak sensitive information?
00:12:59.339 --> 00:13:01.179
Are dependencies maintained?
00:13:01.419 --> 00:13:03.819
Does the application have appropriate tests?
00:13:03.979 --> 00:13:08.699
And perhaps most importantly, does anyone actually understand what has been built?
00:13:08.939 --> 00:13:13.899
The easier software becomes to create, the easier it becomes to create bad software.
00:13:14.139 --> 00:13:15.899
And that is the paradox.
00:13:16.299 --> 00:13:24.139
AI can dramatically reduce the cost of creating software, but it can also dramatically reduce the cost of creating software that nobody understands.
00:13:24.459 --> 00:13:28.859
That is why I don't think the future is simply non-developers replacing developers.
00:13:29.019 --> 00:13:37.819
I think the future is developers becoming dramatically more productive while non-developers gain the ability to create software that previously required a development team.
00:13:38.059 --> 00:13:39.819
Those are two different things.
00:13:40.059 --> 00:13:42.139
And there is another shift taking place.
00:13:42.379 --> 00:13:45.500
We used to think of software development as a linear process.
00:13:45.740 --> 00:13:54.539
A product manager writes a specification, a designer creates the interface, a developer writes the code, a tester tests it, an operations team deploys it.
00:13:54.859 --> 00:13:57.259
AI is beginning to collapse those boundaries.
00:13:57.500 --> 00:14:10.939
One person can describe an idea, an agent can generate the interface, another agent can create the back end, another can write tests, another can investigate security, another can review the code, and the human becomes the orchestrator.
00:14:11.179 --> 00:14:13.019
That is what I mean by vibe tooling.
00:14:13.099 --> 00:14:14.939
It's not just vibe coding.
00:14:15.179 --> 00:14:21.099
It is an idea that natural language becomes a control layer across the entire software development process.
00:14:21.339 --> 00:14:24.299
And that raises a really interesting question.
00:14:24.620 --> 00:14:27.179
What happens to the skill of programming?
00:14:27.419 --> 00:14:31.339
I don't think programmers disappear, but I think the value changes.
00:14:31.579 --> 00:14:33.899
Knowing how to write a loop is useful.
00:14:34.139 --> 00:14:37.659
Knowing how to design a secure system is much more valuable.
00:14:37.819 --> 00:14:44.139
Likewise, knowing syntax is useful, but knowing what architecture you should use is much more valuable.
00:14:44.379 --> 00:14:51.659
The AI can generate an enormous amount of implementation, but someone still has to decide whether that implementation is appropriate.
00:14:51.819 --> 00:14:55.579
That means critical thinking becomes more important, not less.
00:14:55.819 --> 00:15:05.019
Architecture thinking, security and testing all become more important, and the ability to communicate clearly with machines becomes a genuine technical skill.
00:15:05.339 --> 00:15:07.099
Because prompting is evolving.
00:15:07.259 --> 00:15:11.419
The early version of vibe coding was often just build me a website.
00:15:11.659 --> 00:15:13.259
That works for a demonstration.
00:15:13.419 --> 00:15:16.620
It doesn't work particularly well for a complex production system.
00:15:16.939 --> 00:15:25.500
The better approach is to describe the objective, the constraints, the users, the business rules, the architecture, the acceptance criteria and the tests.
00:15:25.740 --> 00:15:30.379
In other words, the better you understand the problem, the better you can direct the AI.
00:15:30.539 --> 00:15:34.379
And perhaps that is the biggest lesson from comparing these four platforms.
00:15:34.620 --> 00:15:39.899
The AI isn't eliminating the need for experience, it's changing where experience matters.
00:15:40.299 --> 00:15:45.259
So if we look ahead, I think the distinction between these products will gradually become less clear.
00:15:45.579 --> 00:16:01.259
Lovable is becoming more developer-friendly, Repli is becoming more autonomous, Cursor will become more accessible to non-developers, GitHub Copilot will become more agentic, and all of them will increasingly use multiple models rather than relying on a single underlying AI model.
00:16:01.419 --> 00:16:03.979
That is another fascinating part of this competition.
00:16:04.219 --> 00:16:06.779
The real differentiator may not be the model.
00:16:06.939 --> 00:16:26.779
It may be the agent harness around the model, the tools the agent can use, the context it can access, the quality of the code-based understanding, the ability to test its own work, the ability to maybe recover from errors, the security controls, the workflow around human approval, and the integration with the rest of the technology ecosystem.
00:16:27.019 --> 00:16:34.620
In other words, the future isn't necessarily about having the smartest AI, it's about giving the AI the right environment in which to work.
00:16:34.779 --> 00:16:37.179
And that takes us back to the four platforms.
00:16:37.339 --> 00:16:52.539
Lovable feels like an AI product builder, Replib feels like an AI development environment, Cursor feels like an AI software engineer sitting inside your code base, and GitHub Copilot increasingly feels like an AI layer sitting across the software development lifecycle.
00:16:52.779 --> 00:16:57.500
None of them are universally better, they simply optimize for different starting points.
00:16:57.740 --> 00:17:04.859
And if I had to give one piece of advice to anyone experimenting with these tools, it would be this don't start by asking which tool is best.
00:17:05.099 --> 00:17:07.660
Start by asking what are you trying to build?
00:17:07.900 --> 00:17:13.259
If you are validating an idea, use the tools that get you from idea to prototype the fastest.
00:17:13.500 --> 00:17:18.059
If you're learning, choose something that lets you see and understand what the AI is doing.
00:17:18.220 --> 00:17:23.819
If you're maintaining a code base, choose a tool that understands your repository and development workflow.
00:17:24.059 --> 00:17:28.940
And if you're running software development at enterprise scale, look beyond coding experience.
00:17:29.099 --> 00:17:32.460
Examine governance, security, integration and control.
00:17:32.940 --> 00:17:37.019
Because the most important thing about Vibe tooling isn't that AI can write code.
00:17:37.180 --> 00:17:38.620
We've known that for a while.
00:17:38.859 --> 00:17:45.419
The really important development is that AI is beginning to take responsibility for much larger parts of the software development process.
00:17:45.659 --> 00:17:57.180
We're moving from autocomplete to agents, from agents to autonomous workflows, and potentially from individual AI assistants to teams of AI agents working alongside human teams.
00:17:57.419 --> 00:18:02.059
That doesn't mean the developer disappears, it means the definition of a developer changes.
00:18:02.299 --> 00:18:13.819
The developer of the future may spend less time writing lines of code and more time deciding on what should be built, how it should be built, whether it is safe, whether it is maintainable, and whether it actually solves the problem.
00:18:13.980 --> 00:18:16.700
And perhaps that is the real promise of vibe tooling.
00:18:16.859 --> 00:18:22.779
Not that we stop building software, but we can spend more time thinking about what the software can actually do.
00:18:23.019 --> 00:18:32.700
However, I suspect we're going to look back at 2026 as an early chapter in this story, because the real transformation isn't going to be about which AI coding tool wins.
00:18:32.859 --> 00:18:36.700
It's going to be about what happens when creating software becomes conversational.
00:18:36.859 --> 00:18:41.339
When the interface to your development environment isn't primarily a keyboard and an IDE.
00:18:41.500 --> 00:18:42.700
It's a conversation.
00:18:42.859 --> 00:18:52.539
You describe the outcome, the AI figures out the implementation, you review it and refine it, and eventually, perhaps, you simply say what you want the software to become.
00:18:52.779 --> 00:19:02.299
That is the world of vibe tooling, and it could fundamentally change who gets to build software, how software is built, and ultimately how quickly ideas can become reality.
00:19:02.539 --> 00:19:03.659
Well that's all for today.
00:19:03.740 --> 00:19:05.980
Thanks for tuning in to the Inspiring Tech Leaders podcast.
00:19:06.059 --> 00:19:10.059
If you've enjoyed this episode, don't forget to subscribe, leave a review, and share it with your network.
00:19:10.220 --> 00:19:15.019
You can find more insights, show notes, and resources at www.inspiringtechleaders.com.
00:19:15.180 --> 00:19:16.619
Head over to the social media channels.
00:19:16.700 --> 00:19:19.819
You can find Inspiring Tech Leaders on X, Instagram, Inspo, and TikTok.
00:19:19.899 --> 00:19:22.139
And let me know your thoughts on vibe tooling.
00:19:22.299 --> 00:19:27.819
Thanks for listening, and until next time, stay curious, stay connected, and keep pushing the boundaries of what's possible in tech.
00:00:00.239 --> 00:00:03.919
Welcome to the Inspiring Tech Leaders podcast with me, Dave Roberts.
00:00:04.160 --> 00:00:09.519
Today I'm going to look at vibe tooling and compare GitHub Copilot, Replic, Cursor, and Lovable.
00:00:09.919 --> 00:00:12.080
The idea of vibe tooling is simple.
00:00:12.320 --> 00:00:20.719
Instead of sitting down and writing thousands of lines of code yourself, you describe what you want to build and an AI system works out how to build it.
00:00:20.960 --> 00:00:29.440
You can write the code, create the files, connect databases, run tests, identify errors, make changes and increasingly deploy the finished application.
00:00:29.679 --> 00:00:33.359
And there are now several serious players competing for this space.
00:00:33.600 --> 00:00:39.359
They all use AI to help us build software, but they approach the problem from very different directions.
00:00:39.600 --> 00:00:46.640
And that distinction is becoming increasingly important because the question is no longer simply which AI coding assistant is best.
00:00:46.880 --> 00:00:54.159
The much more interesting question is what kind of software builder are you, and what sort of relationship do you want to have with the AI?
00:00:54.399 --> 00:00:56.560
Let's start with the term vibe coding.
00:00:56.719 --> 00:01:01.119
The phrase became popular because it captured a very different way of creating software.
00:01:01.280 --> 00:01:10.560
You have an idea, you describe it in natural language, the AI generates the code, you look at what has been produced, tell it what you want changed, and continue iterating.
00:01:10.799 --> 00:01:23.519
You might say, Build me a custom portal, add authentication, give it a dashboard, connect it to a database, add a search function, make the interface look modern, and add in an administration area.
00:01:23.760 --> 00:01:32.640
Historically that would have been a significant software development project, but with today's tools it's becoming a conversation, and that is the important change.
00:01:32.879 --> 00:01:36.400
We are moving from writing software to directing software.
00:01:36.640 --> 00:01:45.280
The human increasingly becomes the person who defines the problem, describes the desired outcome, makes decisions and judges whether the result is good enough.
00:01:45.519 --> 00:01:51.040
The AI is increasingly becoming the implementation engine, but there is a big catch.
00:01:51.280 --> 00:02:01.120
The various different platforms don't give you the same experience, and I think the easiest way to understand them is to imagine them as four different people building the same application.
00:02:01.599 --> 00:02:04.480
Imagine you want to build a simple business application.
00:02:04.640 --> 00:02:13.439
It has a web interface, users need to log in, there's a database, there is an administration dashboard, and there are a few integrations with external services.
00:02:13.680 --> 00:02:19.360
If I give that requirement to Lovable, I'm essentially saying here is the product I want, build it.
00:02:19.680 --> 00:02:27.439
Lovable is particularly focused on taking an idea expressed in natural language and turning it into a working full stack application.
00:02:27.680 --> 00:02:39.680
Its current approach can handle the front-end, back-end authentication, database and integrations, and its agent can plan more complicated tasks, work through multiple steps, and test the application.
00:02:39.919 --> 00:02:45.599
That makes Lovable perhaps the clearest expression of what many people mean when they talk about vibe coding.
00:02:45.759 --> 00:02:51.039
You don't necessarily start by thinking about the technology stack, you start with the product.
00:02:51.280 --> 00:02:54.240
You might say, I want a dashboard for a sales team.
00:02:54.479 --> 00:03:00.400
It needs a login screen, a customer list, sales chart, and an area where managers can see performance.
00:03:00.560 --> 00:03:04.159
And Lovable starts turning that description into an application.
00:03:04.400 --> 00:03:06.159
The appeal is obvious.
00:03:06.400 --> 00:03:18.400
If you're a founder, product manager, designer, business analyst, or technically curious person who doesn't necessarily want to spend hours working through configuration files, frameworks, and dependencies, this is incredibly powerful.
00:03:18.639 --> 00:03:24.960
You can concentrate on what you want the application to do rather than exactly how the applications should be constructed.
00:03:25.120 --> 00:03:28.719
And Lovable has pushed this further during 2026.
00:03:28.879 --> 00:03:37.599
Its autonomous capabilities allow it to take larger pieces of work, plan them, and continue working without requiring you to micromanage every step.
00:03:37.919 --> 00:03:40.479
There is another interesting development here as well.
00:03:40.800 --> 00:03:48.960
Lovable has introduced ways for applications created on the platform to work with AI tools such as ChatGPT and Claude through MCP.
00:03:49.120 --> 00:03:58.319
So we're beginning to see a world where AI simply isn't helping us build applications, AI can also become a way for users to interact with those applications.
00:03:58.560 --> 00:04:00.960
That is quite a significant shift.
00:04:01.280 --> 00:04:03.120
Now let's move to Replit.
00:04:03.360 --> 00:04:05.919
Replit takes a slightly different approach.
00:04:06.240 --> 00:04:16.399
It is still very accessible and very focused on describing what you want and allowing an agent to build it, but it gives you more visibility into the underlying development environment.
00:04:16.720 --> 00:04:25.199
Replit's current agent can build applications from natural language and the platform combines development, execution and deployment in one environment.
00:04:25.439 --> 00:04:37.439
Its Agent 4 release in 2026 emphasises autonomous work, parallel agents, and the ability to deal with things such as authentication, databases, back-end functionality and front-end design.
00:04:37.680 --> 00:04:46.000
So if Lovable feels like you're saying Build me an application, Replit feels more like saying, Build me an application, but let me look under the bonnet.
00:04:46.240 --> 00:04:47.680
That distinction matters.
00:04:48.000 --> 00:04:52.800
Replic can be an excellent bridge between the traditional developer and the non-developer.
00:04:53.360 --> 00:05:00.000
You can start with a natural language prompt, get an application running very quickly, and then start exploring the code.
00:05:00.240 --> 00:05:02.639
You can see what the AI has actually created.
00:05:02.800 --> 00:05:08.720
You can experiment, you can run it and change it as needed, and perhaps most importantly, you can learn from it.
00:05:08.879 --> 00:05:17.199
That makes Repli particularly interesting for people who want the speed of vibe coding without completely abandoning their traditional software development experience.
00:05:17.439 --> 00:05:19.920
Imagine someone who has an idea for an application.
00:05:20.079 --> 00:05:28.160
They may not know Python or JavaScript, they may not understand databases, they may not know what an API is, but they can describe their idea.
00:05:28.399 --> 00:05:32.639
The AI creates something, they see it working, they start asking questions.
00:05:32.879 --> 00:05:34.240
What does this file do?
00:05:34.399 --> 00:05:36.240
Why have you created this database?
00:05:36.480 --> 00:05:38.399
What happens if I change this?
00:05:38.720 --> 00:05:42.160
That is a completely different way of learning software development.
00:05:42.399 --> 00:05:50.240
Instead of learning syntax first and building something later, you can build something first and then learn the underlying technology as you go.
00:05:50.480 --> 00:05:55.279
And that's why I think Replit sits somewhere between vibe coding and traditional development.
00:06:34.139 --> 00:06:35.740
Now we come to Cursor.
00:06:35.980 --> 00:06:41.819
Cursor is probably the tool in this comparison that is closest to the traditional professional software developer.
00:06:42.139 --> 00:06:52.780
Cursor is an AI-powered development environment built around the idea that AI should understand your existing code base and become an agent capable of making substantial changes across it.
00:06:53.100 --> 00:07:05.020
Cursor describes itself as a coding agent for building ambitious software, and its current platform supports autonomous agents, code-based understanding, multiple models, cloud agents, and parallel development.
00:07:05.259 --> 00:07:07.500
And this is an important distinction.
00:07:07.819 --> 00:07:16.620
Cursor doesn't primarily say give me an idea and I'll create your application from scratch, it says give me your software and let me help you develop it.
00:07:16.860 --> 00:07:24.220
Of course, Cursor can absolutely build new applications, but its real strength becomes apparent when you already have a serious code base.
00:07:24.780 --> 00:07:42.540
You might have millions of lines of code, you might have a complex React application, you might have a Python backend, you might have hundreds of tests, you might also have authentication, APIs, databases, and deployment pipelines, and you can ask Cursor to understand that environment and then make a significant change.
00:07:42.780 --> 00:07:55.579
For example, you could say add multi-factor authentication to the application, update the login flow, modify the database schema, update the tests and make sure the existing authentication behaviour isn't broken.
00:07:55.980 --> 00:08:00.060
This is a very different proposition from simply generating a landing page.
00:08:00.300 --> 00:08:03.980
Cursor has already been pushing heavily into autonomous agents.
00:08:04.220 --> 00:08:15.579
Its cloud agents can operate in isolated environments, make changes, test the software, and produce artifacts such as screenshots, videos and logs so that a developer can review what happened.
00:08:15.819 --> 00:08:18.939
And this is where the term vibe tooling becomes useful.
00:08:19.180 --> 00:08:25.100
Because the human is no longer necessarily sitting there watching the AI type, you can give an agent a task.
00:08:25.259 --> 00:08:29.020
It can go away and work on that task and then come back with the results.
00:08:29.180 --> 00:08:31.980
You review the changes and decide whether to accept them.
00:08:32.299 --> 00:08:39.740
Cursor has been moving towards fleets of agents working in parallel, where different agents can work on different pieces of the code base.
00:08:39.980 --> 00:08:46.139
That begins to look less like an AI autocomplete tool and more like an AI software engineering team.
00:08:46.379 --> 00:08:48.860
Now let's talk about GitHub Copilot.
00:08:49.259 --> 00:08:56.939
This is where the comparison becomes particularly interesting because GitHub Copilot has evolved enormously from the product many people originally knew.
00:08:57.100 --> 00:09:01.419
The old model of GitHub Copilot was essentially an AI pair programmer.
00:09:01.579 --> 00:09:05.179
You start typing the code and it suggests the next piece of code.
00:09:05.340 --> 00:09:06.939
You accept it and continue.
00:09:07.179 --> 00:09:11.259
That is still part of GitHub Copilot, but it's no longer the whole story.
00:09:11.419 --> 00:09:21.659
GitHub now has copilot agent mode inside development environments where the agent can determine which files need to be changed, execute commands, iterate and remediate problems.
00:09:21.819 --> 00:09:24.379
And then there is GitHub Copilot Cloud Agent.
00:09:24.699 --> 00:09:27.979
This is where GitHub's strategy becomes particularly powerful.
00:09:28.219 --> 00:09:33.500
You can give GitHub Copilot a task and let it work in the background in its own development environment.
00:09:33.659 --> 00:09:41.659
It can investigate the repository, make changes, run tests, and ultimately produce a pull request or leave the work on a branch for you to review.
00:09:41.979 --> 00:09:47.019
So imagine you have a GitHub repository with an issue saying improve the search experience.
00:09:47.259 --> 00:09:53.979
Previously, a developer would read this issue, investigate the code, make changes, run tests, and create the pull request.
00:09:54.139 --> 00:09:57.339
Now you can increasingly assign that task to an agent.
00:09:57.579 --> 00:10:01.099
The agent investigates, plans, writes code and tests.
00:10:01.259 --> 00:10:03.019
The human then reviews.
00:10:03.339 --> 00:10:06.859
That is a fundamental change in the software development workflow.
00:10:07.099 --> 00:10:16.939
GitHub has also added things such as model selection, self-review, security scanning, custom agents, and CLI handoff to its coding agent experience.
00:10:17.259 --> 00:10:21.899
And this brings us to what I think is the most important difference between the four tools.
00:10:22.219 --> 00:10:25.979
Lovable is extremely focused on turning an idea into an application.
00:10:26.219 --> 00:10:31.500
Repli is focused on making that process accessible while keeping the development environment visible.
00:10:31.740 --> 00:10:35.659
Cursor is focused on making an existing code base highly agentic.
00:10:35.819 --> 00:10:43.819
GitHub Copilot is increasingly focused on putting AI agents directly into the software development's lifecycle and GitHub workflow.
00:10:44.059 --> 00:10:45.579
So which one is best?
00:10:45.819 --> 00:10:47.979
Well, the answer is it depends.
00:10:48.219 --> 00:10:55.019
For example, if I'm a non-technical founder and I have an idea for a SaaS product, I would probably start by looking at lovable.
00:10:55.339 --> 00:10:59.099
I don't necessarily want to understand the architecture on day one.
00:10:59.339 --> 00:11:07.259
I want to see the product experiment and then have the ability to say, that's not quite right, change this, add that, or remove this.
00:11:07.500 --> 00:11:10.139
That's where lovable becomes extremely compelling.
00:11:10.379 --> 00:11:15.899
If I want more visibility into the code and the development environments, Replic becomes very attractive.
00:11:16.059 --> 00:11:17.819
It's a great middle ground.
00:11:18.139 --> 00:11:22.859
It allows you to move more quickly while still giving you access to the underlying machinery.
00:11:23.099 --> 00:11:29.019
If I'm a professional developer working on an existing software product, Cursor becomes extremely compelling.
00:11:29.259 --> 00:11:37.740
The ability to understand a code base, work across multiple files, use different models, and delegate substantial engineering tasks is very powerful.
00:11:37.979 --> 00:11:49.500
And if my organisation is already heavily invested in GitHub, GitHub actions, pull requests, issues, security controls, and enterprise deployment processes, GitHub Copilots has a huge advantage.
00:11:49.740 --> 00:12:00.059
It isn't just an AI coding tool, it sits in the software development ecosystem, and that matters, because enterprise software development isn't simply about generating code.
00:12:00.219 --> 00:12:07.259
It's about governance, security, identity, source control, code review, testing, auditability, and deployment.
00:12:07.500 --> 00:12:12.859
It's about understanding who changed what, why they changed it and whether that change is safe.
00:12:13.099 --> 00:12:18.699
This is one of the reasons I think the enterprise battle around AI coding is going to be particularly fascinating.
00:12:18.939 --> 00:12:22.859
The winner won't necessarily be the tool that writes the best piece of code.
00:12:23.099 --> 00:12:26.859
It could be the platform that creates the best overall development system.
00:12:27.099 --> 00:12:30.139
And that brings us to one of the dangers of vibe coding.
00:12:30.379 --> 00:12:34.939
Just because an AI can build something doesn't mean the thing is built is good.
00:12:35.179 --> 00:12:38.539
This is probably the biggest misconception surrounding the whole movement.
00:12:38.779 --> 00:12:41.019
You can ask an AI to build an application.
00:12:41.179 --> 00:12:45.979
It might work, it might look fantastic, it might even have thousands of lines of code behind it.
00:12:46.139 --> 00:12:48.139
But does it have a sensible architecture?
00:12:48.299 --> 00:12:49.259
Is it secure?
00:12:49.419 --> 00:12:51.099
Does it handle edge cases?
00:12:51.259 --> 00:12:53.339
Are permissions correctly implemented?
00:12:53.579 --> 00:12:55.659
Is the database designed properly?
00:12:55.899 --> 00:12:56.859
Can it scale?
00:12:57.099 --> 00:12:59.019
Does it leak sensitive information?
00:12:59.339 --> 00:13:01.179
Are dependencies maintained?
00:13:01.419 --> 00:13:03.819
Does the application have appropriate tests?
00:13:03.979 --> 00:13:08.699
And perhaps most importantly, does anyone actually understand what has been built?
00:13:08.939 --> 00:13:13.899
The easier software becomes to create, the easier it becomes to create bad software.
00:13:14.139 --> 00:13:15.899
And that is the paradox.
00:13:16.299 --> 00:13:24.139
AI can dramatically reduce the cost of creating software, but it can also dramatically reduce the cost of creating software that nobody understands.
00:13:24.459 --> 00:13:28.859
That is why I don't think the future is simply non-developers replacing developers.
00:13:29.019 --> 00:13:37.819
I think the future is developers becoming dramatically more productive while non-developers gain the ability to create software that previously required a development team.
00:13:38.059 --> 00:13:39.819
Those are two different things.
00:13:40.059 --> 00:13:42.139
And there is another shift taking place.
00:13:42.379 --> 00:13:45.500
We used to think of software development as a linear process.
00:13:45.740 --> 00:13:54.539
A product manager writes a specification, a designer creates the interface, a developer writes the code, a tester tests it, an operations team deploys it.
00:13:54.859 --> 00:13:57.259
AI is beginning to collapse those boundaries.
00:13:57.500 --> 00:14:10.939
One person can describe an idea, an agent can generate the interface, another agent can create the back end, another can write tests, another can investigate security, another can review the code, and the human becomes the orchestrator.
00:14:11.179 --> 00:14:13.019
That is what I mean by vibe tooling.
00:14:13.099 --> 00:14:14.939
It's not just vibe coding.
00:14:15.179 --> 00:14:21.099
It is an idea that natural language becomes a control layer across the entire software development process.
00:14:21.339 --> 00:14:24.299
And that raises a really interesting question.
00:14:24.620 --> 00:14:27.179
What happens to the skill of programming?
00:14:27.419 --> 00:14:31.339
I don't think programmers disappear, but I think the value changes.
00:14:31.579 --> 00:14:33.899
Knowing how to write a loop is useful.
00:14:34.139 --> 00:14:37.659
Knowing how to design a secure system is much more valuable.
00:14:37.819 --> 00:14:44.139
Likewise, knowing syntax is useful, but knowing what architecture you should use is much more valuable.
00:14:44.379 --> 00:14:51.659
The AI can generate an enormous amount of implementation, but someone still has to decide whether that implementation is appropriate.
00:14:51.819 --> 00:14:55.579
That means critical thinking becomes more important, not less.
00:14:55.819 --> 00:15:05.019
Architecture thinking, security and testing all become more important, and the ability to communicate clearly with machines becomes a genuine technical skill.
00:15:05.339 --> 00:15:07.099
Because prompting is evolving.
00:15:07.259 --> 00:15:11.419
The early version of vibe coding was often just build me a website.
00:15:11.659 --> 00:15:13.259
That works for a demonstration.
00:15:13.419 --> 00:15:16.620
It doesn't work particularly well for a complex production system.
00:15:16.939 --> 00:15:25.500
The better approach is to describe the objective, the constraints, the users, the business rules, the architecture, the acceptance criteria and the tests.
00:15:25.740 --> 00:15:30.379
In other words, the better you understand the problem, the better you can direct the AI.
00:15:30.539 --> 00:15:34.379
And perhaps that is the biggest lesson from comparing these four platforms.
00:15:34.620 --> 00:15:39.899
The AI isn't eliminating the need for experience, it's changing where experience matters.
00:15:40.299 --> 00:15:45.259
So if we look ahead, I think the distinction between these products will gradually become less clear.
00:15:45.579 --> 00:16:01.259
Lovable is becoming more developer-friendly, Repli is becoming more autonomous, Cursor will become more accessible to non-developers, GitHub Copilot will become more agentic, and all of them will increasingly use multiple models rather than relying on a single underlying AI model.
00:16:01.419 --> 00:16:03.979
That is another fascinating part of this competition.
00:16:04.219 --> 00:16:06.779
The real differentiator may not be the model.
00:16:06.939 --> 00:16:26.779
It may be the agent harness around the model, the tools the agent can use, the context it can access, the quality of the code-based understanding, the ability to test its own work, the ability to maybe recover from errors, the security controls, the workflow around human approval, and the integration with the rest of the technology ecosystem.
00:16:27.019 --> 00:16:34.620
In other words, the future isn't necessarily about having the smartest AI, it's about giving the AI the right environment in which to work.
00:16:34.779 --> 00:16:37.179
And that takes us back to the four platforms.
00:16:37.339 --> 00:16:52.539
Lovable feels like an AI product builder, Replib feels like an AI development environment, Cursor feels like an AI software engineer sitting inside your code base, and GitHub Copilot increasingly feels like an AI layer sitting across the software development lifecycle.
00:16:52.779 --> 00:16:57.500
None of them are universally better, they simply optimize for different starting points.
00:16:57.740 --> 00:17:04.859
And if I had to give one piece of advice to anyone experimenting with these tools, it would be this don't start by asking which tool is best.
00:17:05.099 --> 00:17:07.660
Start by asking what are you trying to build?
00:17:07.900 --> 00:17:13.259
If you are validating an idea, use the tools that get you from idea to prototype the fastest.
00:17:13.500 --> 00:17:18.059
If you're learning, choose something that lets you see and understand what the AI is doing.
00:17:18.220 --> 00:17:23.819
If you're maintaining a code base, choose a tool that understands your repository and development workflow.
00:17:24.059 --> 00:17:28.940
And if you're running software development at enterprise scale, look beyond coding experience.
00:17:29.099 --> 00:17:32.460
Examine governance, security, integration and control.
00:17:32.940 --> 00:17:37.019
Because the most important thing about Vibe tooling isn't that AI can write code.
00:17:37.180 --> 00:17:38.620
We've known that for a while.
00:17:38.859 --> 00:17:45.419
The really important development is that AI is beginning to take responsibility for much larger parts of the software development process.
00:17:45.659 --> 00:17:57.180
We're moving from autocomplete to agents, from agents to autonomous workflows, and potentially from individual AI assistants to teams of AI agents working alongside human teams.
00:17:57.419 --> 00:18:02.059
That doesn't mean the developer disappears, it means the definition of a developer changes.
00:18:02.299 --> 00:18:13.819
The developer of the future may spend less time writing lines of code and more time deciding on what should be built, how it should be built, whether it is safe, whether it is maintainable, and whether it actually solves the problem.
00:18:13.980 --> 00:18:16.700
And perhaps that is the real promise of vibe tooling.
00:18:16.859 --> 00:18:22.779
Not that we stop building software, but we can spend more time thinking about what the software can actually do.
00:18:23.019 --> 00:18:32.700
However, I suspect we're going to look back at 2026 as an early chapter in this story, because the real transformation isn't going to be about which AI coding tool wins.
00:18:32.859 --> 00:18:36.700
It's going to be about what happens when creating software becomes conversational.
00:18:36.859 --> 00:18:41.339
When the interface to your development environment isn't primarily a keyboard and an IDE.
00:18:41.500 --> 00:18:42.700
It's a conversation.
00:18:42.859 --> 00:18:52.539
You describe the outcome, the AI figures out the implementation, you review it and refine it, and eventually, perhaps, you simply say what you want the software to become.
00:18:52.779 --> 00:19:02.299
That is the world of vibe tooling, and it could fundamentally change who gets to build software, how software is built, and ultimately how quickly ideas can become reality.
00:19:02.539 --> 00:19:03.659
Well that's all for today.
00:19:03.740 --> 00:19:05.980
Thanks for tuning in to the Inspiring Tech Leaders podcast.
00:19:06.059 --> 00:19:10.059
If you've enjoyed this episode, don't forget to subscribe, leave a review, and share it with your network.
00:19:10.220 --> 00:19:15.019
You can find more insights, show notes, and resources at www.inspiringtechleaders.com.
00:19:15.180 --> 00:19:16.619
Head over to the social media channels.
00:19:16.700 --> 00:19:19.819
You can find Inspiring Tech Leaders on X, Instagram, Inspo, and TikTok.
00:19:19.899 --> 00:19:22.139
And let me know your thoughts on vibe tooling.
00:19:22.299 --> 00:19:27.819
Thanks for listening, and until next time, stay curious, stay connected, and keep pushing the boundaries of what's possible in tech.
