11 Comments
User's avatar
David Wilkens's avatar

Temporal is going to have it's moment.

Ashu Joshi's avatar

+1 on Temporal, they are the ones to watch.

Roxane Googin's avatar

Yes but, isn't including including multiple probabilistic steps in what one imagines will be a deterministic outcome likely to get dicey? How do you account for the inherent variability of multiple small probabilistic decisions made from changing data on the fly as you chain them together? What becomes the "right" answer? Who judges? You could end up building little Towers of Babel instead of useful processes.

Giridhar Vishwanath's avatar

Workflows are the new databases, and workflows are exactly where the ROI of AI will become visible. It's one thing to say "30% of our code is written by AI" - feel good about it, without having to show any definite ROI - and a better thing to say "10 out of 40 workflows are driven by AI". But, with AI, will these workflows remain the same? With Gen AI becoming better with each model, the workflow to reach the end result may become shorter. And what tools will be standard for designing these workflows? Will it be the incumbents like AgentForce and SalesForce, maybe something similar in ServiceNow, SAP, Workday - because this is where all the workflows are.

Harshit Shah's avatar

'Workflows are the new databases' really resonated with me as it captures how the real impact of AI isn’t just in doing a task well, but in reorganizing the whole system around that task.

Ref: https://platforms.substack.com/p/heath-ledger-and-the-untold-half

Jason Pohl's avatar

Jamin, you just made the case for Databricks. Databricks Apps provide a seamless way to host applications and takes care of the Authorization for the developer. By using the Python SDK, the developer has access to Databricks Workflows, Lakebase, Model Serving, Agent Bricks, Genie (Text-to-SQL), Data Warehousing, and every other component on the Databricks platform that is needed to build a compound AI Application.

Hugo Lu's avatar

Finally people realising that Agentic workflows require orchestrators

95% of an "agent" is work that can be handled by serverless, quick to start orchestrators doing what is fundamentally data work.

Johnny b's avatar

Been hearing a lot about this new cybersecurity company "Cyera" - any interest in covering them in a future blog?

Matt Carbonara's avatar

Temporal and Orkes should benefit from this trend

Ashu Joshi's avatar

This is an excellent analysis. I looked at the topic from a slightly different angle - the need for orchestration- and posted about it on my Substack as well. https://movinture.substack.com/p/agentic-ai-and-role-of-orchestration

David Wilkens's avatar

Temporal is going to have it's moment.