When initially experimenting with LLMs and agentic AI, software program engineers at Notion AI utilized superior code era, advanced schemas, and heavy instructioning.
Shortly, although, trial and error taught the group that it might do away with all of that sophisticated knowledge modeling. Notion’s AI engineering lead Ryan Nystrom and his group pivoted to easy prompts, human-readable representations, minimal abstraction, and acquainted markdown codecs. The end result was dramatically improved mannequin efficiency.
Making use of this re-wired method, the AI-native firm launched V3 of its productiveness software program in September. Its notable function: Cutomizable AI brokers — which have shortly turn into Notion’s most profitable AI instrument thus far. Based mostly on utilization patterns in comparison with earlier variations, Nystrom calls it a “step operate enchancment.”
“It's that feeling of when the product is being pulled out of you relatively than you making an attempt to push,” Nystrom explains in a VB Past the Pilot podcast. “We knew from that second, actually early on, that we had one thing. Now it's, ‘How might I ever use Notion with out this function?’”
‘Rewiring’ for the AI agent period
As a conventional software program engineer, Nystrom was used to “extraordinarily deterministic” experiences. However a lightweight bulb second got here when a colleague suggested him to easily describe his AI immediate as he would to a human, relatively than codify guidelines of how brokers ought to behave in numerous eventualities. The rationale: LLMs are designed to know, “see” and motive about content material the identical approach people can.
“Now, at any time when I'm working with AI, I’ll reread the prompts and power descriptions and [ask myself] is that this one thing I might give to an individual with no context and so they might perceive what's occurring?” Nystrom stated on the podcast. “If not, it's going to do a foul job.”
Stepping again from “fairly sophisticated rendering” of knowledge inside Notion (reminiscent of JSON or XML) Nystrom and his group represented Notion pages as markdown, the favored device-agnostic markup language that defines construction and that means utilizing plain textual content with out the necessity for HTML tags or formal editors. This enables the mannequin to work together with, learn, search and make modifications to textual content recordsdata.
In the end, this required Notion to rewire its techniques, with Nystrom’s group focusing largely on the middleware transition layer.
In addition they recognized early on the significance of exercising restraint in terms of context. It’s tempting to load as a lot data right into a mannequin as doable, however that may gradual issues down and confuse the mannequin. For Notion, Nystrom described a 100,000 to 150,000 token restrict because the “candy spot.”
“There are circumstances the place you may load tons and tons of content material into your context window and the mannequin will battle,” he stated. “The extra you place into the context window, you do see a degradation in efficiency, latency, and in addition accuracy.”
A spartan method can also be vital within the case of tooling; this may help groups keep away from the “slippery slope” of countless options, Nystrom suggested. Notion focuses on a “curated menu” of instruments relatively than a voluminous Cheesecake Manufacturing unit-like menu that creates a paradox of selection for customers.
“When folks ask for brand new options, we might simply add a instrument to the mannequin or the agent,” he stated. However, “the extra instruments we add, the extra selections the mannequin has to make.”
The underside line: Channel the mannequin. Use APIs the way in which they have been meant for use. Don't attempt to be fancy, don't attempt to overcomplicate it. Use plain English.
Hearken to the total podcast to listen to about:
Why AI remains to be within the pre-Blackberry, pre-iPhone period;
The significance of "dogfooding" in product growth;
Why you shouldn’t fear about how value efficient your AI function is within the early phases — that may be optimized later;
How engineering groups can maintain instruments minimal within the age of MCP;
Notion’s evolution from wikis to full-blown AI assistants.
Subscribe to Past the Pilot on Apple Podcasts, Spotify, and YouTube.
[/gpt3]
