An AI project that drifts for months is frustrating and expensive. Before blaming the engineer, it helps to ask: did we hire the wrong person, or did we ask the right person to do the wrong job?
Signs the scope is wrong
- No clear success metric. Nobody can say what "done" looks like.
- The goal keeps changing. Every demo leads to a new direction.
- The data is not there. The engineer spends weeks finding and cleaning data.
- Research when you needed delivery. You asked for a working tool but the role was framed as "explore AI".
Signs the hire is wrong
- Only notebooks, never products. Experiments look good but never reach users.
- Poor communication. Progress is hard to understand and updates are rare.
- Over-engineering. Custom models are built where an existing API would do.
- Weak software basics. No version control, tests or documentation.
- Missed small commitments — a reliable early indicator of bigger problems.
A quick diagnosis
| Symptom | Likely cause |
|---|---|
| Great prototypes, nothing in production | Hired a researcher for an engineering job |
| Constant rework after demos | Unclear or changing scope |
| Slow progress, vague updates | Skills or communication gap |
| Costs rising with no end date | No milestones or success metric |
How to recover
- Pause and rescope. Define one measurable outcome for the next four weeks.
- Review what exists. Have an independent developer assess the code and data.
- Set weekly demos of working software, not slides.
- Match the skill set to the work. Integration and automation work suits a strong applied developer.
- Secure your assets. Make sure you own repositories, accounts and documentation.
Prevention is simpler
Start with a short, paid pilot, insist on production-ready deliverables and keep milestones small. Most AI hiring problems are avoided with clear scope and regular demos.
Need a second opinion on a stalled project? Ask our team for a code and scope review.