Welcome to the latest edition of Dura Digital’s newsletter. This month, we’re exploring the innovations reshaping how teams build, design, and collaborate.
Welcome to the latest edition of Dura Digital’s newsletter. This month, we’re exploring the innovations reshaping how teams build, design, and collaborate. You’ll find insights on advocating for user research, Microsoft’s move to expand flexibility in its AI strategy, and how teams can turn the estimation process from a sticking point into a strength. We also highlight a new approach that enables AI models to forget sensitive data, opening the door to more responsible AI systems.
Bridging the estimation gap
Early project estimates often feel risky because teams are asked to commit before all the details are known. The key to moving forward isn’t forcing precision but creating safety and clarity, using high-level estimation techniques to set direction and reinforcing that estimates are provisional, not promises. By shifting the approach, teams can unlock funding, build trust, and keep projects moving with confidence.
Many organizations still skip user research, often citing excuses like cost, time, or fear of negative feedback. The reality is that skipping this step leads to wasted effort, poor product decisions, and missed opportunities to truly understand user needs. Shifting the mindset and treating research as a fast, cost-effective tool for clarity can transform both products and the culture around building them.
Microsoft continues to add flexibility to its AI future
Microsoft has introduced its first in-house AI models, signalling a new phase in its strategy for Copilot. The MAI-Voice-1 model delivers high-speed, high-fidelity audio generation, while MAI-1-preview is being tested for text-based use cases. Together, they highlight Microsoft’s move to build alternatives alongside OpenAI, giving the company more flexibility and control over its AI future.
A breakthrough in forcing AI to forget private data
A new approach from UC Riverside researchers shows how AI models can be made to “forget” private or copyrighted data, even without access to the original training datasets. By using surrogate data and carefully calibrated noise, the method erases targeted information while prese rving the model’s functionality. This could help organizations comply with privacy laws, protect sensitive data, and reduce the costly need for full retraining.
