{"id":14617,"date":"2026-10-21T11:08:09","date_gmt":"2026-10-21T09:08:09","guid":{"rendered":"https:\/\/say-digital.io\/blog\/ai-speeds-up-coding-projects-still-slow\/"},"modified":"2026-10-21T11:08:22","modified_gmt":"2026-10-21T09:08:22","slug":"ai-speeds-up-coding-projects-still-slow","status":"publish","type":"post","link":"https:\/\/say-digital.io\/blog\/ai-speeds-up-coding-projects-still-slow\/?lang=en","title":{"rendered":"AI speeds up coding. Why are your projects still slow?"},"content":{"rendered":"
\n

AI speeds up coding. Why are your projects still slow?<\/h1>\n

The feature is built. It still cannot ship: the business owner has not settled an exception, test data is missing and approval is waiting for the next meeting. The developer saved time. The project is still waiting.<\/p>\n

AI can accelerate code production without shortening a project by the same amount.<\/strong> As implementation gets faster, scoping, dependencies and acceptance can become the limiting steps. For a business, the useful measure is elapsed time from an identified need to an outcome people can actually use, with comparable scope and quality.<\/p>\n

Where does the time go outside coding?<\/h2>\n

An internal tool does not start with the first commit or finish with the last generated line. Someone must decide what users need to do, obtain access, settle business rules, test exceptions and authorise the release. Some tasks run in parallel. Others prevent everything downstream from moving.<\/p>\n

If the team is waiting for a pricing decision, building the billing screen faster will not settle it. If the approver is available only once a week, sending more work for review may lengthen the queue. Faster implementation does not remove that dependency; it exposes it.<\/p>\n

This does not explain every delay. A difficult integration or a risky migration may still be the main technical constraint. Trace the work through the actual project before deciding that the problem must be organisational.<\/p>\n

What the research supports \u2014 and what it does not<\/h2>\n

The 2024 DORA report<\/a> identifies individual benefits from AI adoption alongside a negative association with delivery stability and throughput. It highlights small batches and robust testing. These survey findings do not prove that AI slows every team; they show why individual gains cannot reliably predict delivery performance.<\/p>\n

Measurement is changing, too. In its 24 February 2026 update<\/a>, METR explains why its newer experiment produces an unreliable signal: some developers do not want to work without AI, some tasks are excluded through selection, and using several agents makes time tracking harder. METR considers it likely that developers are getting more acceleration than in early 2025, but cannot establish the size of that increase confidently.<\/p>\n

Neither those studies nor a successful demo tells you when your next project will ship. They are reasons to measure your own delivery process, not apply a coding-productivity percentage to the entire schedule.<\/p>\n

The code is ready. The business rule is not.<\/h2>\n

An illustrative scenario, not a measured client case.<\/em> A small business wants to automate overdue-invoice reminders. AI helps build the interface and draft the messages. Testing then raises three questions: should disputed invoices receive reminders? Who can pause a message? Which system is authoritative for payment status?<\/p>\n

Each answer changes how the software behaves. If the answers arrive after implementation, the team revises the rules, screens and tests. Time saved on the first version now pays for some of the rework instead of bringing the release closer.<\/p>\n

Better scoping would not require weeks of exhaustive documentation. It would bring the collections owner, someone who understands the data and the developer together around representative invoices. They would agree on an initial scope, release-blocking exceptions and the person who accepts the result. A focused PRD and design sprint<\/a> can turn those decisions into build instructions; a polished mockup with unstated rules cannot.<\/p>\n

Shorten the waits without removing the controls<\/h2>\n

The wrong response is to remove approvals so the business can \u201ckeep up with AI\u201d. An incorrect customer message or accounting entry costs more than a timely decision. Make the controls specific and available rather than deleting them.<\/p>\n