Super-Convenient Software, or The End of Workflows as We Know Them

As the founder of a fledgling tech startup, I ponder at length almost on a daily basis, how AI may impact software in general, and enterprise software in particular.

While we know AGI is out there to be discovered in time, we have business to develop and run in the meantime. How are we going to evolve teams, organizations, and products to serve customers best?

After giving it considerable thought, I’ve developed a framework to help me identify opportunities and act on them, keeping the customer at the forefront of decision-making.

The Bane of Enterprise Software: Workflows

Much of enterprise software is about setting up, maintaining, and updating rigid workflows:

1. The HR/Finance department uses payroll software to keep track of various payroll-related tasks that must be executed in a particular sequence (semi-automated).
2. Developers set up a Docker and CI/CD test and build mechanism for delivering software updates, which must run on every major change.
3. Sales teams use CRM systems to track progress in turning prospects with varying interest levels into customers.

The key characteristic of many of these enterprise processes is simple: they contain a massive number of steps to execute. A significant amount of internal resources is allocated first to design the given workflow. Then, even more time and energy are dedicated to training people to follow the workflow. As the business evolves, the process must be iterated upon to keep the workflows current and relevant.

Thus, complex and cost-intensive workflows are the bane of corporate existence. They involve:

1. Design costs
2. Setup costs
3. Training costs
4. Maintenance costs
5. Update costs

Like delicate flower gardens, enterprise workflows also require a huge amount of attention, care, and resources to make them profitable.

A more concrete & personal example

At our organization, we manage nearly a dozen microservices, which translates to handling hundreds of APIs.

Despite being an advocate for high-standards engineering, my team and I personally procrastinated on integrating Swagger or OpenAPI into our infrastructure. The setup costs of incorporating a standard API documentation tool like Swagger involve the following steps for each API repository we maintain:

1. Check whether there is Swagger support for the repository's language/framework.
2. Add the extra dependency to the project and test it thoroughly.
3. Set up the CI system to generate documentation on every update.
4. Create a URL endpoint for team access.
5. Manage deployment and hosting properly.
6. Handle SSL for the endpoints.
7. Set up Apache or another proxy if you want the "Try" feature to work.

These steps, in sum, could easily take a few days to up to a week for a newbie engineer to get going. Moreover, we have to replicate the same setup for almost every different microservice backend. And it so happens that we take advantage of multiple languages (JS/Python/Go) and various frameworks within these languages. Just finding the right Swagger code generators for each language/framework is enough of a headache to compel us to procrastinate on the task.

Looking at the sheer amount of setup cost, as well as the estimated maintenance costs, we found some additional effort-saving mechanisms of our own. In fact, we invented our own text-oriented language for dealing with this issue. It sort of solved the problem for us, but the usability still wasn’t great either—we wanted a better solution.

The Impending Death of Workflows

Traditional software has usually involved many "steps" to get things done.

In an AI-first world, what we will see is a "compression" of steps.

What used to take 8-10 steps can be reduced to 1 step—or no steps at all.

Therefore, the key point with AI for enterprise software is exponential convenience.

Wherever we see a workflow with multiple steps, we can try to invent AI-enabled components to bring it down to 1 step.

What used to require human intervention can now be solved with zero human input to a large extent (with good review/feedback mechanisms from humans).

Case in point: with the Swagger/OpenAPI example, we ultimately ended up building LiveAPI, which essentially reduces the "setup cost per repository" to "2 clicks."

Moreover, due to the generality of underlying LLMs, our tool supports most of the popular languages/frameworks out of the box. We also perform language/framework detection automatically, along with automated base URL detection and other similar features.

Essentially, the whole Swagger workflow—previously involving considerable setup and maintenance cost—has been reduced to a "set and forget" kind of system, requiring just a few clicks for setup.

The Dawn of Super-Convenient Software

For entrepreneurs, the goal in the near term could very well be to focus on producing all sorts of Super-Convenient Software.

The defining property of an Super-Convenient Software System is the total compression of an existing multi-step workflow.

You take something with many steps and compress it into a simple single-step onboarding process. Thereafter, the AI-powered system continues to deliver value—with almost no human intervention required.

The vision of Super-Convenient Software opens up many doors for innovation. At Hexmos, we are proud to have started our efforts with LiveAPI, which can reduce weeks' worth of effort in traditional workflow-based systems into just a few minutes.