Mark Herring has spent more than three decades selling things that resist being sold in a sentence. Java platforms. Hadoop. Time-series databases. Secure infrastructure access. Data versioning. Industrial messaging. Now enterprise AI. Each arrived with its own architecture diagram, devoted developers and a small blizzard of acronyms. Herring's job has been to find the simple human proposition hiding inside the technical truth, then make it earn its keep.
This is why his arrival at H2O.ai in 2026 feels less like a career swerve than the next station on a very consistent railway. H2O.ai builds predictive, generative and agentic AI for organizations that care where the models run, who controls them and how their decisions are governed. The vocabulary is new. The adoption puzzle is familiar: engineers can see the capability before the market understands the consequence.
Herring tends to meet that puzzle with two instruments. One is language. Strip away the brochure varnish, listen to the customer and say plainly what changes. The other is measurement. Count the journey from first attention to qualified opportunity, find the place where it leaks, and refuse to confuse motion with progress. If his work has a recurring protagonist, it is the developer. If it has a recurring antagonist, it is the attractive metric that cannot pay an invoice.
The category before the category
His corporate itinerary reads like an annotated history of how software became infrastructure. He began at Oracle in the early 1990s, moved to Forte Software, and joined Sun Microsystems after Sun acquired Forte in 1999. At Sun, his responsibilities touched Java, web services, software infrastructure and MySQL. At Software AG, he worked in senior product leadership as integration, real-time systems and adaptive applications moved toward the center of enterprise computing.
Then came the open-source chapters. At Hortonworks, Herring led corporate and developer marketing around Hadoop. When InfluxData recruited him as CMO in 2017, its announcement credited him with growing the Hortonworks developer community by more than forty times. At InfluxData, the hard thing to explain was time itself: why machines, sensors and applications produced data with needs a general-purpose database did not neatly meet.
He helped give that explanation a pleasing stopwatch. “Time to awesome” was the distance between a developer's first download and the first useful result. The phrase is cheerful, but the idea is severe. Every obstacle in installation, onboarding or early use taxes adoption. Developer experience is therefore part product design, part marketing and part clock.
One career, six technical markets
Count the pipe, not the confetti
In a long CXOTalk conversation about the CMO's craft, Herring called marketing a competition for eyeballs and mindset. Then he made the important turn: attention must lead somewhere. His teams hold themselves accountable to pipeline, a measure that sales, finance and the board can all debate in the same currency. Leads and page views remain useful signals. They lose their crown.
This stance makes marketing less of a performance staged for the rest of the company and more of a jointly operated system. Herring described his relationship with sales as two people driving the same car. The metaphor is slightly alarming, but the organizational idea is sound: marketing accepts a number it cannot create alone, sales accepts responsibility for the opportunities it takes, and both sides inspect the route.
The habit beneath it all is instrumentation. Herring has joked about always wanting another date-and-time stamp. He studies cohorts, touchpoints and conversion paths, looking for a pattern competitors missed. In his recent writing, the same instinct appears as a funnel diagnostic: identify which leads correlate with revenue, locate the stage where conversion collapses, and make sales and marketing agree on what “qualified” means before buying more demand.
“Become a doer and let data be your resume.”Mark Herring, advice to marketers
The line captures both his optimism and his impatience. Experiments may fail. A marketer should still be bold enough to run them and disciplined enough to bring back evidence. Herring's own advice to people who want the CMO job begins with doing, not theorizing. The promotion case is a result someone else can inspect.
A developer, a marketer and a free shirt
Data alone would make for a dry lunch companion. Herring's other instrument is an amused understanding of how marketing looks from the outside. He once recalled chatting with a developer who announced that he hated marketing people. Herring pointed out that he was one. The developer allowed an exception because Herring had given him a shirt.
The joke works because both parties recognize the bargain. Developers dislike being smothered in adjectives, and marketers sometimes arrive bearing cotton diplomacy. Herring's deeper response is to describe products through customers, acknowledge what the software does poorly and create content people actually need. “Content without SEO doesn't exist,” he has said, a characteristically unsentimental reminder that a useful answer nobody finds remains lonely.
At trade shows, everyone on his team has done booth duty, including Herring. The point is more than egalitarian theatre. Tough questions test whether a story works in the wild. They expose which sentence requires a diagram, which claim earns an eye roll and which problem makes a buyer lean closer. The booth is a live editorial desk with worse carpeting.
When the workflow gets a memory
Herring's more recent work turns the instrument back on the marketing organization. His test for an AI-native company is concrete: remove the AI tools tomorrow and see what stops. If people merely lose a quicker way to draft an email, the operating model remains intact. A consequential implementation changes scoring, routing, research, context or execution.
He has been building that change into his own routine. One system keeps a written record of his working context: relationships, companies, open threads and calendar commitments. It reviews that memory before helping him decide where to focus. The division of labor is deliberate. The system prepares drafts and surfaces the queue; Herring retains decisions about whom to answer, what to say and what to spend.
Cost supplied its own comic lesson. In 2026, he wrote that he burned a month of personal AI budget in three days by using an expensive model for routine work and failing to cap subagent tokens. His fix was routing: reserve costly reasoning for jobs that need it, send routine work elsewhere and put a ceiling on the meter. The episode sounds like cloud-finance history wearing a new lanyard. Cheap units encourage abundant use. Someone still needs to own the bill.
His team design follows the same logic. H2O.ai's biography describes a preference for a few skilled people running optimized systems with agents and agentic workflows. This is less a forecast of empty offices than a demand for operational clarity. A task should belong to a person, an agent or a purposeful collaboration between them. “Everyone sort of uses AI” is not an operating chart.
The H2O.ai chapter
Herring joined H2O.ai with a compact thesis: enterprise AI is moving from experiment to production. In production, model quality shares the stage with ownership, governance, deployment, security, observability and cost. A dazzling demo can ignore those matters for ten minutes. A regulated company cannot ignore them on Monday morning.
This gives him another technically respected product whose wider story must be made legible. H2O.ai has open-source roots and a long record in machine learning. Herring's task is to connect that credibility to the current buyer's anxiety: how to run predictive, generative and agentic systems where the business lives, with controls the organization can defend.
The aspiration sounds grand only until Herring starts drawing the plumbing. Find the leak. Clarify the buyer. Route the work. Measure the spend. Put the right content where the technical audience already gathers. Let customer experience do more talking than adjectives. Then make the resulting pipeline visible enough that sales and marketing can argue about the same fact.
After Java, Hadoop, time-series data and industrial messaging, AI presents an expansive vocabulary for Herring to translate. His advantage is a stubbornly small set of questions. Does the buyer understand it? Can a developer use it? Does the workflow change? Did it create a result? Software markets reinvent themselves with impressive frequency. The useful questions age rather well.