When developing internal AI tools, adopt a 'fail fast' mantra. Many use cases fail not because the idea is bad, but because the underlying models aren't yet capable. It's critical to regularly revisit these failed projects, as rapid advancements in AI can quickly make a previously unfeasible idea viable.

Avoid vague, company-wide AI mandates. Instead, apply a maturity framework to individual processes (e.g., account research). This approach builds a practical roadmap, moving specific use cases up the maturity ladder as needed and preventing costly over-engineering.

Initial failure is normal for enterprise AI agents because they are not just plug-and-play models. ROI is achieved by treating AI as an entire system that requires iteration across models, data, workflows, and user experience. Expecting an out-of-the-box solution to work perfectly is a recipe for disappointment.

Ambitious AI projects may fail their primary goal but still produce valuable secondary assets. An attempt to predict memory prices with an LLM failed, but the automated data gathering process created a first-of-its-kind historical analysis dashboard, which proved to be a more valuable outcome.

The OpenAI Codex app would have "absolutely failed" if launched three months earlier. The only difference was the underlying model's capability. This reveals a new product risk: a perfectly designed product can fail simply because the AI isn't smart enough yet, requiring teams to relaunch ideas as models improve.

The 85% AI project failure rate isn't a technology problem. It stems from four business and process issues: failing to identify a narrow use case, using data that isn't clean or ready, not defining success and risk, and applying deterministic Agile methods to probabilistic AI development.

Many AI projects become expensive experiments because companies treat AI as a trendy add-on to existing systems rather than fundamentally re-evaluating the underlying business processes and organizational readiness. This leads to issues like hallucinations and incomplete tasks, turning potential assets into costly failures.

A great source for high-impact AI projects is your company's 'graveyard' of past initiatives. Revisit projects that were strategically sound but failed because they were too time-consuming or administratively burdensome. The manual effort that made them unfeasible is often what AI is best suited to automate now.

Building AI applications is a moving target. Engineering solutions to compensate for current model deficiencies (like limited context windows) is often wasted effort, as future models will likely solve those problems. The key is to anticipate the capabilities of the model you'll have at launch and not bet against its progress.