AI can build the software. Why is implementing it still so expensive?

01

Enterprise requirements rarely exist in isolation

Consider a seemingly straightforward requirement: “We want to improve our employee onboarding process.” At first glance, this sounds like an HR problem. Implementing it properly requires answers across the business:

An experienced consultant may recognise these dependencies immediately. Someone less familiar with the organisation may discover them only after implementation has begun. Neither the initial request nor the code itself necessarily captures the full business context — which is why enterprise projects become complicated even when the individual technical components are relatively straightforward.

02

The hidden cost of implementation is often coordination

Implementation work involves much more than development. People need to explain requirements, clarify assumptions, make decisions, resolve disagreements, review documentation, and coordinate dependencies — frequently across meetings, emails, messaging applications, and project management tools.

And if coordination remains expensive, the overall cost of implementation may not fall as quickly as the cost of generating code.

03

Why a more capable AI assistant is not necessarily the complete answer

General-purpose AI assistants can work with documents, interact with connected applications, generate code, summarise conversations, and execute multi-step tasks. Any serious enterprise AI strategy should take that into account. But there is a distinction between an assistant that can perform a task and a system that reliably manages a business process.

Suppose an assistant reads an email describing a requirement and creates a task in a project management system. That is useful. These questions remain:

These questions concern the state and integrity of the implementation process, not just the ability to execute individual tasks. A general-purpose assistant may be configured to address many of them. The opportunity for specialised implementation software is to make this behaviour systematic, repeatable, and verifiable. We are not assuming general-purpose AI cannot perform these activities — we are asking whether implementation-specific knowledge, structured state, and explicit validation can make the overall process more reliable and economical.

04

From conversations to implementation-ready requirements

One direction we are exploring is to help teams capture requirements through the channels they already use. People should not need to learn a new interface or complete an elaborate form every time they need to explain a business requirement. A stakeholder might describe it through a conversation or voice message, and the system helps turn it into structured project context:

The system should distinguish what was explicitly stated from what remains to be clarified. It should not silently turn every inference into an approved requirement. Once the necessary information is confirmed, the structured requirement can be recorded in the appropriate project management or implementation system. The objective is not more documentation for its own sake — it is reducing the effort required to move from an informal conversation to an agreed, actionable requirement.

05

Why ERP-specific knowledge matters

Enterprise software is not a blank canvas. Platforms such as Odoo, SAP, and ERPNext already contain business functionality, data models, permissions, workflows, and transaction rules. Rebuilding these capabilities unnecessarily would undermine much of the value of using an ERP in the first place.

A useful implementation system should therefore understand the capabilities and constraints of the underlying ERP: what can be handled through standard configuration, what requires additional development, and what needs human review. The objective is not to replace the ERP. It is to make the ERP easier to implement correctly.

06

The opportunity for implementation partners

The conversation also raised a commercial question: how can implementation businesses increase revenue or reduce costs? We believe there is an opportunity to help partners automate parts of their existing delivery process without forcing them to replace their established systems.

Imagine a partner receiving a new customer enquiry. Today the team might research the customer, collect requirements, identify relevant modules, prepare discovery documents, follow up on missing information, and transfer the scope into its project management system. An implementation operations layer could help:

The partner would retain its customer relationships, delivery expertise, and control over important decisions. If the approach works, a partner could spend less time on administrative coordination and more time on consulting — and a small delivery team could handle more opportunities. These are hypotheses to validate, not outcomes we can claim without measurement.

07

Start with a narrow workflow, not an entire autonomous enterprise

It is tempting to automate the entire implementation lifecycle at once. That would be a mistake. Discovery, architecture, configuration, development, migration, testing, training, deployment, and support each carry different risks. A more practical approach is a narrow workflow where the current process is expensive, repetitive, and measurable — for example, converting customer enquiries into structured discovery briefs.

The relevant measures are concrete:

Does the workflow reduce the time required to prepare a discovery brief? Does it identify missing requirements earlier? Does it introduce incorrect assumptions? These are questions that can be answered with evidence. If the workflow produces measurable value, it can be extended into implementation planning, documentation, verification, and delivery coordination.

08

What we are exploring at WorkSimplified

Connecting business intent to the work required to deliver it.

We are exploring how AI can help capture requirements, preserve decisions, identify dependencies, generate implementation-ready documentation, and maintain continuity across the people and systems involved — fitting into the tools implementation partners already use, rather than requiring them to abandon their existing processes.

The first wave of AI adoption demonstrated that natural language can be an effective interface to software. The next challenge is making these capabilities dependable within the complex processes businesses already operate: understanding requirements, maintaining context, managing dependencies, and verifying outcomes — while recognising where human expertise remains essential.

We still need to establish where this approach creates the greatest value, how much work can be automated safely, and whether the savings justify the cost of adoption. Those questions should be answered through real implementations, not assumptions. The goal is not to eliminate implementation expertise. It is to help that expertise go further. At WorkSimplified, that is the problem we are working to solve.