Welcome to the Proactive Bandwagon
Deconstructing the new AI buzzword - "Proactive AI"
It appears we’ve arrived at the inevitable conclusion. The buzzword of the moment in AI is ‘proactive AI’.
OpenAI launched Pulse, a feature where ChatGPT does research based on your personal context while you sleep, as a new ‘proactive paradigm for interacting with AI’. The idea is that you’ll have a feed of potentially helpful information waiting for you when you wake up (sounds a bit like the business model of Indian IT services promising 24/7 work when the US time zone sleeps).
If you’ve been following Babbage, you’d know that we’ve been hammering on the idea of proactive AI for a while. It’s nice to see the rest of the world join this conversation.
Now that the word is out there in public, we expect it to be thrown around liberally for business-facing AI software.
What the hell is proactive AI?
New words slip out of the AI hype thesaurus every so often, and like ‘agentic’, we can expect a fair bit of confusion and obfuscation with this new buzzword too.
The basic idea of a proactive AI involves ‘anticipatory cognition.’ The AI does not wait for the user to initiate an action. Since so many initial AI launches were reactive co-pilot chatbots, this indicates a shift to the opposite modality, where the AI no longer waits for the user to initiate an action. Essentially, the locus of control moves from “human prompts → AI responses” to “AI hypotheses → human verification.”
At least, that’s the ideal definition.
However, it’s all a bit of a word soup at the moment. Consider this linguistic Möbius strip used to describe Microsoft Copilot: “specialized, proactive Copilot Agents.“ It makes one wonder what anything means.
To simplify things, let’s ignore Copilot as an idea altogether. But what’s the difference between agentic and proactive?
Agentic vs. Proactive
A simplified argument could be that all “Proactive AI” is Agentic (in the sense it’s doing something by itself), but not all Agentic AI is proactive.
Many of last year’s “AI Agents” were automated workflow connectors. SaaS agents automated human workflows by sending context from one place to the next. Browser-based consumer agentic capabilities (like Perplexity’s Comet or Agent mode in ChatGPT) pushed buttons and filled forms. So essentially, Agentic AI that wasn’t proactive.
Proactive AI, on the other hand, is your CRM deciding when to follow up with a lead or your consumer agent writing a birthday message to your close friend. Pulse is proactive because it conducts research on behalf of the user without the need for the user to initiate it.
This ‘doing’ will be the most marketed and flashy part of proactive AI, but the place where we will start seeing true leverage is further up the chain - the thinking.
There’s a proactive AI for that
While all Proactive AI involves thinking to some extent, it is helpful to distinguish which parts of the thinking-doing cycle the AI is proactive about.
Any outcome has three components to it:
Proactive Execution Agent: Think of them like your digital interns. Think of Zendesk automatically responding to customer queries. Quack, which recently received $7 million in seed funding, explicitly states that it builds “proactive AI agents that behave like customer service reps.”
It’s an AI agent with an external signal or trigger that ‘acts’ proactively. If you strip away complexity enough this becomes robotic process automation but maybe with a chirpy chat bot.
Proactive Decision Agent: Here, the proactivity extends to making choices. A simple proactive AI that knows what to say to a customer query isn’t making much of a choice, whereas some of the Agentic SaaS products do straddle a decision layer. When Salesforce AI suggests when to follow up or when a deal is stalling, it makes some business decisions. It could then potentially call upon Salesforce agents to execute them.
Proactive Insight Engine: This is where the proactive thinking happens. Isn’t every proactive AI doing some thinking? For example, Pulse proactively ‘thinks’ about what would be useful to the user. However, the bar gets higher if we want to elevate proactive thinking to proactive insights.
If we look at it through that lens, Pulse is more like a personalized news feed, and its proactivity weighs heavily on the execution side of the proactive agent spectrum.
Proactive thinking accelerators
The closer a proactive agent is to the execution side of the spectrum, the more it’s about “saving money”. It’s essentially an “output accelerator.” Fewer humans or multiply output with the same number of humans.
On the other hand, the closer a proactive agent is to the insight side of the spectrum, the more its about “better decisions.” The selling point now becomes finding new opportunities or making step changes in your business. It’s a different level of leverage.
There’s a lot of value to be extracted all along this spectrum, as we’re still in the early stages of reconfiguring our economic engine.
However, the real breakthroughs in proactivity will likely come from proactive insight engines that can truly generate valuable insights. These insights will lead to high-impact decisions, which will, in turn, inform the execution agents that are available to carry them out.
The powerful version of “proactive AI” will build companies that can think for themselves.
~Babbage Insight