Welcome to 2025! It’ll be the Yr of AI Brokers in martech, in phrase and deed. Right here’s what I believe which means…
From the second there was a couple of software program product in an organization’s tech stack, there’s been jockeying to find out who was boss.
If the tech stack had been an orchestra, which product could be the conductor?
Whichever one directed the move of information amongst apps and decided which perform by which app was referred to as when — actually orchestrating the tech stack — would wield a place of nice energy.
This is the reason it was a strategic error for advertising and marketing suites of the early 2010’s to downplay integrations with different merchandise within the burgeoning martech panorama. They pretended different merchandise weren’t vital or necessary. To have strived to be the conductor would have been to acknowledge the worth of these different apps. They didn’t wish to concede that time.
As Julia Roberts’ character in Fairly Girl chided, “Large mistake. Large. Big!”
Suites’ failure to grab the baton created a gap for an entire new class of merchandise generally known as “iPaaS” — integration-platform-as-a-service — to step as much as the rostrum and construct multi-billion greenback corporations, similar to Mulesoft, Workato, and Zapier. iPaaS embraced the range of tech stacks and empowered individuals to automate digital operations at a degree above particular person merchandise.
The suites lastly realized that they had been turning into a spoke on the sting of the stack graph as a substitute of the hub at its heart. Many switched to platform methods and pursued constructing app ecosystems. Those that didn’t slipped into irrelevance. Salesforce acquired Mulesoft. HubSpot acquired PieSync.
However the barn door had already been open, and the cows by no means absolutely got here house. To this present day, iPaaS merchandise — a lot of which have advanced into “enterprise automation platforms” — retain a robust presence in most advertising and marketing tech stacks. They’re co-conductors of the orchestra. For some use instances, the advertising and marketing platform calls the pictures. For others, the enterprise automation platform née iPaaS does.
Mockingly, historical past almost repeated itself a number of years later. CDPs — buyer knowledge platforms — arose between 2015-2020 as a result of most advertising and marketing suites/platforms had been nonetheless not open sufficient or versatile sufficient with their knowledge fashions. This gave CDPs an iPaaS-like opening, they usually stampeded by it.
However the advertising and marketing suites/platforms weren’t as resistant to reply this time. Most acquired a CDP or constructed CDP-like capabilities by the 2020’s. (Some overcompensated by buying or constructing a number of CDPs, however that’s somewhat inside baseball.) In the meantime, in a shock twist, cloud knowledge warehouses/lakehouses disrupted the entire CDP landscape. That’s a narrative for an additional day although.
Which brings us to 2025. The fashionable martech stack has a number of platforms vying to be the conductor on the “heart” of all of it. Many advertising and marketing ops groups attempt to standardize on one to automate workflows throughout their toolset, though which platform they select — CRM, MAP, CDP, CDW, and so on. — varies from firm to firm. However almost 1/3 additionally use an iPaaS/enterprise automation platform too.
Advanced? Considerably. Manageable? Roughly.
However brace your self. It’s about to get waaaay extra advanced with — drum roll — AI brokers.
Whereas there’s plenty of debate about what qualifies as a “true” AI agent — particularly round how a lot decision-making autonomy it has — one attribute that the majority brokers share: instrument use. They will work together with different software program, both by APIs or by simulating a human consumer’s clicks and typing, in an RPA-like trend. They apply such instrument use to learn or write knowledge from totally different sources, take actions in different apps, and even make requests to different brokers. The latter is the imaginative and prescient of multi-agent networks.
AI brokers can tackle duties, processes, or much more open-ended targets. They automate digital operations throughout totally different merchandise in your tech stack in an built-in trend.
Apart from the goal-seeking twist, that sounds acquainted, no?
AI brokers reverberate with the echoes of iPaaS and workflow automation platforms. In actual fact, most of the distributors in these classes have remodeled into AI agent constructing platforms. The distinction? Workflow automations ran in line with deterministic guidelines and steps that had been programmed by a human. AI brokers apply their very own intelligence to resolve which instruments to name when — with out an specific plan dictated to them by people.
Current martech SaaS platforms aren’t dragging their ft to the AI agent get together both. Salesforce has swung their narrative pendulum all the best way to Agentforce. HubSpot has its Breeze AI brokers, and co-founder Dharmesh Shah launched agent.ai, “the skilled community for AI brokers.” Microsoft has integrated Copilot Studio for constructing AI brokers in Dynamics. Braze, Bloomreach, and Zeta have introduced their very own AI brokers too.
Now, if we had been merely coping with current iPaaS/automation distributors and ecosystem-oriented SaaS platforms morphing into AI agent platforms, that might be difficult sufficient. Within the brief time period, digital ops will turn into a tangle of previous deterministic workflow automations intermingled with new clever agent automations. We’ll need to migrate to new methods of architecting advertising and marketing operations processes accordingly. However we’ll get the upside of brokers doing extra work autonomously, decreasing the reams of guidelines maintained by hand. (Hallelujah, proper?)
Nevertheless, current SaaS and iPaaS/automation corporations are not the one gamers within the pit. A number of new classes of AI brokers have joined in, every bringing their very own tackle clever orchestration of particular person productiveness and/or organizational course of automation.
Right here’s a fast tour of the opposite agent contributors within the map above…
The core AI engines from Anthropic, Google, Meta, OpenAI, et al., all help agent improvement, with “perform calling” for instrument use as a primary primitive. Anthropic’s not too long ago introduced “laptop use” functionality and Mannequin Context Protocol are thrilling steps ahead in foundational AI agent capabilities.
Agent developer frameworks similar to Autogen, AutoGPT, Camel AI, Griptape, LangChain, and LlamaIndex all assist facilitate corporations constructing their very own customized brokers. For those who can code with APIs — and with AI-powered developer instruments similar to Cursor.so, GitHub Copilot, and Replit, more and more anybody can code — you possibly can construct your personal tailor-made AI brokers in a matter of days or hours.
Need much more assist in constructing your personal AI brokers? There’s an entire new class of no-code/low-code agent builders: CrewAI, Lindy, Relevance.AI, Sema4.ai, SmythOS, and Tektonic.AI, to call only a few. The boundary between these and advanced iPaaS/workflow automation platforms is extraordinarily fuzzy. I’d merge them, besides it’s price mentioning what number of new native AI agent builder platforms have emerged on this class.
On the far different finish of the spectrum, Anthropic, Google, Microsoft, and OpenAI are all incorporating instrument use of their end-user merchandise similar to Claude, ChatGPT, Copilot, and Gemini. More and more, utilizing Gemini inside Google Workspace or Copilot in Microsoft 365 merchandise similar to Excel and Groups will carry out agent-like automations for people. There will likely be growing overlap between private productiveness and enterprise course of automation.
Unbiased browser-based AI automation instruments similar to Bardeen and Axiom.ai will likely be part of the combo as nicely. No matter web-based software program a consumer has entry to, they will create AI brokers (or pseudo-agents) to automate their work with these apps — even when the distributors of these merchandise have carried out nothing expressly to allow that.
There’s a particular class of brokers which have emerged to leverage reinforcement studying — a sub-branch of machine studying — to autonomously optimize personalised advertising and marketing campaigns and applications. OfferFit is a pioneer on this house (disclosure: I’ve been an advisor to them). Aampe and composable CDP Hightouch have additionally entered the class, which is coming to be generally known as “AI Decisioning.”
Final however positively not least, there’s a large wave of role-specific AI agent instruments, particularly for gross sales and customer support. Consultant AI brokers for gross sales embody 11x, AISDR, Artisan, Momentum, regie.ai, and Rox. Buyer success AI brokers embody Decagon, MavenAGI, and Sierra.ai. In fact, current SaaS platforms are competing with their very own role-specific brokers, similar to these from HubSpot, Intercom, Salesforce, and Zendesk.
The factor all these totally different AI brokers have in widespread?
They’re all automating interactions — more and more autonomously — with the merchandise and platforms in your stack. Every thinks it’s orchestrating its piece of the symphony. However will there be one conductor that sees throughout your entire orchestra and harmonizes all these gamers? Anticipate intense vendor competitors for that place.
Within the meantime, orchestrating the orchestrators would be the massive problem of Large Ops, in 2025… and for the remainder of the last decade forward.