Vibe coding for business value

What vibe coding is

Vibe coding is software development where AI writes most or all of the code from a described outcome or an intent. The person directing it says what the software should do and leaves the how to the AI agents or platform. This allows subject-matter experts who own the process to create what they always dreamed about.

Depending on what this application is built for and for whom, you may need to take different approaches.

1. 1
   Build without review when the cost of being wrong is low

   Examples are a prototype, a tool that five people use, or a page that is taken down after a conference. If one of these breaks, the cost is small.

   To decide which rule applies, ask who would be harmed if the application gave a wrong result.

2. 2
   Require review when the cost of failure is high

   Review is needed when an application handles personal data or payments, runs at scale, or becomes the place where a figure the business relies on is kept, which is true of many applications that stay in use.

The four categories

1. 1
   Rapid prototyping

   A team builds a working version, puts it in front of users within a week, uses it to decide a question the team disagrees about, and then deletes it.

2. 2
   Data synthesis

   A person asks one question in plain language, and the application answers it from several systems that each hold part of the answer. It reads both the records and the documents stored beside them, and each answer shows the evidence it is based on.

3. 3
   Internal tools

   The applications a function needs to run its own process, built by the people who run it.

   One intake form for a central shared-service team had sat unchanged on SharePoint for five years. The person who maintained it had left, nobody was funded to replace it, and it sat in the IT backlog behind everything with a revenue number attached. A subject-matter expert in that team rebuilt it in three days, and added AI features the original never had.

4. 4
   Client-facing AI applications

   The customers and internal teams you support use these applications directly. Examples are an assessment someone fills in before they first speak to you, and a portal they use after they sign up.

How do these application types connect to the AI Value paths

Each category serves one or more of the three AI value paths. Consider which type to use depending on what you’re trying to achieve.

What rapid building does not solve

Rapid building removes development as the bottleneck. The value then depends on how well you decide what to build, model the data and get people to adopt it.

1. 1
   Deciding what deserves to exist

   When building software was slow and expensive, the cost of development limited what got built. Now that a team can build ten applications in the time one used to take, the limit is the team’s judgment about which ones to build, and sometimes the right decision is to build none of the ten.

2. 2
   Modeling the data

   Applications built quickly often run into problems with their data model. The intake form described above worked because the person who rebuilt it already knew what information a request contains.

3. 3
   Getting people to adopt it

   An application that people do not use brings no value, however little it cost to build. Rapid building also increases the number of applications competing for the same people’s time and attention.

How to get this right

The people building are subject-matter experts without software engineering training. Set guardrails, principles and guidelines before anyone starts building, because adding them after many applications exist means migrating each application to the new rules.

• Decide who can see what before you populate the app with any data.
• Read data from the system that owns it instead of copying it in.
• Name the point at which a prototype becomes something people depend on, and review it as one when it gets there.