Maarten Verwaest

The through-line

Across the LinkedIn posts, one line of argument keeps resurfacing: technology, whether AI, a model, or infrastructure, is not the point of arrival. The workflow around it is. "It's the workflow, stupid." [3] and "the finish line isn’t when it compiles" [1] frame nearly every product post, from AAF handovers [7][14] to hybrid storage [2]. Over time this product-level conviction widens into a broader economic argument: the same distrust of quick technological fixes shows up in his skepticism toward "AI jobs apocalypse" alarmism [4] and in his critique of BigTech-BigMedia consolidation, where scale and access, not workflow, are what get concentrated [5][12].

On workflow over the model

The recurring claim is that AI and infrastructure investments fail unless they are embedded in the actual production process. "AI does not fix a fragmented workflow. If you add it to chaos, it will multiply the chaos." [10] He argues "the next breakthrough will not come from another model, but from properly embedding AI into the production workflow" [10], and repeatedly calls out tools that stop short: "This is where most allegedly 'AI-powered' workflows quietly hand the problem back to people." [7] The alternative he describes is "an intelligent workflow that identifies highlights, prepares the pre-cut and hands over an edit-ready, multi-track timeline directly to Avid" [14], which he ties back to the same principle stated flatly elsewhere: "the finish line isn’t when it compiles" [1].

On hybrid architecture

On infrastructure, he rejects the on-prem-versus-cloud framing outright: "There should be no hard choice between on-prem and cloud storage." [2] Instead, "proper architecture combines both to balance cost, performance and accessibility depending on the specifics of the workflow" [2], with the goal that "media and processing live where they make most sense, while control, automation and reporting remain central" [2]. The stated payoff is operational, not architectural for its own sake: keeping infrastructure "connected without accepting prohibitive storage and or egress cost" [2].

On market power and consolidation

Beyond product concerns, he applies the same suspicion of concentration to the media-tech landscape. Commenting on Hollywood consolidation, he writes that "competition is gradually being replaced by concentration, access by gatekeeping, and entrepreneurship by rent extraction" [5], and that "the effects of the collusion between BigTech and BigMedia are far stretching" [5]. On the proposed Paramount-Warner Bros merger, he is blunter about what he sees underneath the competitive rationale: "it doesn't like competition between BigMedia and BigTech; more like a gradual alignment of interests" [12], concluding it's "certainly becoming very cosy at the top" [12], while "producers and post-production facilities are expected to deliver more content, faster and with fewer resources" [12]. His underlying standard: "markets only work when power remains contestable and the winners are prevented from closing the door behind them" [5].

On AI and jobs

He pushes back against blanket protection of existing jobs from automation, calling it "'degrowth' dressed up as compassion" [4], and warns that heavy state intervention "usually produces its own unintended averse consequences - the famous cobra effect" [4]. At the same time he does not dismiss the disruption: "We must not ignore disruption. A serious discussion about universal basic income, wage insuran[ce]" [4] is where the thought is left open rather than resolved.

From the stage

In interview settings he goes further into how the workflow philosophy translates into product and company strategy than the LinkedIn posts do. On differentiation, he describes Limecraft as production-oriented workflow software built on turnkey templates rather than a static archive system, integrating with existing on-premises infrastructure through a hybrid model instead of forcing uploads to the cloud [16]. On talent, he explains that Limecraft competes for AI and machine-learning hires against industries like fintech by leaning on the appeal of the media sector itself, using its work with top producers as a recruiting draw, and notes that demand for data science and AI talent currently outstrips supply [17]. On the mechanics of media intelligence, he describes incorporating pre-existing production data, such as casting documents and production reports, into AI engines to improve face and speech recognition accuracy, and insists this requires ongoing human oversight, an iterative process where journalists, documentarians and transcriptionists continuously guide the AI, rather than relying on unattended machine learning or generative AI [18].

Takeaways

• Treat AI output as only half the job: a workflow must still deliver edit-ready timelines, track layouts, markers and comments rather than a pile to reorganise manually [7][14].
• Design infrastructure as hybrid by default, not as an on-prem-versus-cloud choice, to control storage and egress cost while keeping control and reporting centralized [2][16].
• Keep humans in the loop for AI accuracy: face and speech recognition improve when journalists, documentarians and transcriptionists iteratively correct the system rather than leaving it unattended [18].
• Recruit AI and data science talent by leaning on sector appeal (media over fintech) and flagship client work, since supply of such talent lags demand [17].
• Be wary of both extremes on AI-driven job disruption: neither blanket job protection nor ignoring the disruption altogether is presented as sound governance [4].
• Watch for creeping alignment between BigTech and BigMedia interests in major mergers, since it can look like consolidation dressed as competitive necessity [12][5].