The starting point: a substantial build.
A conventional supplier estimated approximately 2,500 human engineering hours to deliver a new application. CEADE took a different approach, using coding agents for much of the implementation and concentrating human work on defining what the application needed to do.
The team had direct access to the client's senior operator, who understood the work the application would support. It was a new build, with no legacy-system integration to work around.
Most human time went into working out the solution.
About 80% of human effort went into problem framing, specifications, scenario design and accountability. People worked through the need, the rules and the situations the application would have to handle, while agents did much of the coding.
This is the thinking behind Signal Logic Factory. We challenge the requirements, work through the user journeys and develop technical specifications before dividing the build into tasks. Clear decisions give the implementation a firm starting point and give reviewers something concrete to test.
Around 200 human hours, in just over a month.
The paper reports delivery in just over a month, using around 200 human hours. That is about 92% below the supplier's quoted hours, or a 12.5:1 comparison.
Reducing implementation effort changes what a delivery team can take on. It also leaves a commercial question: how should the price reflect the human judgement, AI running costs and value of the finished application? That is the wider question the paper explores.
Apply the workflow to the project in front of you.
The Factory can start with a new application, an existing system to improve or replace, or designs and specifications already in hand. Each needs a clear account of what must change, what must stay and how the result will be checked.
CEADE uses the Factory for internal and client projects. Signal Logic Labs can also support an existing development team or work directly with an organisation on its project.
We examine integration needs, access to people who understand the work, security and day-to-day operation when planning delivery. The current Factory uses Codex to run agent work, with the project requirements and reviews organised above that execution environment.
Keep people responsible for the decisions.
The case shows human attention moving towards the work that shapes the application. The Factory carries that principle into its workflow: people decide the scope, acceptable risk and readiness for release; agents carry out tasks within that authority.
Design reviews follow users through complete tasks, including errors and recovery. Usability checks ask whether someone new to the application can complete the work without coaching. Tests follow the requirements and are revisited when those requirements change.
Source-handling rules exclude known secrets. AI usage records keep usage details separate from prompts, responses and personal data.
For applications affecting employees, requirements and reviews need to address changing roles, staff involvement, bias, monitoring and ways to challenge decisions. Those are questions to resolve in the design, alongside the productivity goals.
Make room for work that matters.
A lower delivery effort can make a previously unaffordable project worth considering. Good specifications also bring missing rules, difficult handoffs and operating problems into view before they become expensive changes to the build.
We bring those two concerns together. The aim is a useful application, well thought through and practical to deliver, rather than simply more code produced faster.
Source and context.
Pricing AI Software Services: Reallocating Value and Risk is a working paper by SSL founder Pawel Podolski, written 27 May 2026 and posted on SSRN on 11 June 2026. He is also co-founder of CEADE Solutions, which delivered the engagement described in the paper.
Reading the figures
The supplier quote is an estimate; the 200 hours are the author's account of actual human effort. The comparison concerns human hours. The paper supplies no equivalent quoted calendar duration or verified client-cost saving, so the 12.5:1 ratio applies to effort only.
The application delivery preceded the current evolving Factory workflow. The case explains experience behind the approach; the workflow sections above describe how the Factory organises projects today.
