Date & time

17:00

AI models are changing faster than engineering workflows can absorb. Though major labs ship upgrades every few weeks, teams’ greatest return on investment comes not from frontier-level features but from habits and processes that translate from one model to the next.  It isn’t only the big vendor upgrades that create these challenges; they resurface any time a team tweaks a prompt, adjusts a temperature setting, or changes the weights on a model they’re already running.

Engineering leaders need a structured approach to manage this rate of change. They must evaluate which tools are worth the switch, build an abstraction layer to shield the application from changes, factor in time to test and understand the model, and adapt review processes. Done well, frontier models will take on more complex engineering work. 

This panel brings together engineering leaders who’ve configured systems to ensure they can withstand rapid model upgrades. They’ll explain how to keep your workflows stable while the model underneath it keeps changing, how to test and roll out a new model without a leap of faith, and how to ensure model upgrades save teams time.

Key takeaways:

• Build an abstraction layer that separates your application from any single model
• Test and evaluate a new model before committing to its use
• Adapt your review process so it catches the new bugs a model swap introduces