Kristian Pletten has been exploring AI development since 2014 and today, as the AI lead at Schibsted Partnerstudio, he is harnessing the power of generative AI to reinvent how the studio functions.

During this week’s Webinar, he shared how his team has leveraged AI to extend its reach and reimagine not just how the studio produces work but how it collaborates internally and continues stepping into the future.

Schibsted Partnerstudio is the company’s in-house content marketing agency, producing branded content across Schibsted’s newspapers and digital platforms. Although content marketing articles remain the division’s largest source of revenue, the studio’s staff of 70 now delivers everything from social media campaigns and podcasts to games and feature videos.

The birth of an AI task force

Pletten joined the studio in January 2023, just months after ChatGPT’s release. “For me, this was super exciting,” he said. “I had the perfect job to test out how we could use something like ChatGPT.”

Within weeks, the studio formed an AI task force. Their first step was simple but foundational: Give everyone access to the models. Then came training — understanding things like how large language models work, prompt engineering, the importance of context, and the pitfalls to avoid.

When ChatGPT introduced GPTs, the team organised an AI hack day. Everyone learned to build their own GPTs, marking the studio’s first step toward customised tools.

“But we quickly learned that GPTs had their limits and we wanted to do more. So our next step was to build customised AI toolbox,” Pletten said.   

The toolbox included tools for proofreading, headline generation, meeting transcription, and more. Pletten wrote the code himself — a first for him — using AI coding models.

But even with a growing suite of tools, the workflow remained mostly manual, “with just a touch of AI here and there.” As the capabilities of LLMs improved, he said, “We saw the opportunity to turn this on its head and have most of the workflow become automated, using human guidance and verification at critical stages to create value.”

The goal, from the beginning, was to use AI to gain efficiency and reinvest the time saved into quality, creativity, and customer relationships — not to cut costs.

Producing a single article required roughly 18 hours of work and two weeks from brief to publication, something that Pletten explained: “… works for large campaigns, but not for small repeatable ones.” And with 6,600 clients spending under US$10,000 annually — a segment that continues to grow — the studio needed a way to scale without sacrificing quality.

Cutting time and costs, not quality

The studio’s first major platform was ContentGo, an AI‑driven production system designed to cut production time in half without compromising quality. It automated such tasks as generating briefs and meeting agendas, conducting background research, transcribing interviews, extracting key quotes, and producing first drafts.

At the same time, it preserved the human elements the studio valued: creative meetings, client interviews, and editorial judgment.

Although the results were dramatic, a new challenge emerged: “It turned out that while our ambitious production platform worked, it was cumbersome to use for our content producers. We had mapped out our production workflow, but it turns out that most campaigns are not the same,” Pletten said.

Finding their Flow

This realisation led to the next evolution: Partnerstudio Flow, built entirely from scratch with an agentic architecture.

“An AI agent doesn’t just answer questions,” Pletten explained. “It acts with purpose and takes multiple steps to complete a task.”

Flow provides a more flexible environment where content producers collaborate with specialised agents. Each agent can access tools, gather information, and autonomously work toward a goal.

Flow is more than a tool, Pletten noted; it’s a work methodology covering the entire value chain, from sales to upselling. “We’re moving away from thinking of these systems as tools we use toward making them an integrated part of our daily work.”

The impact was immediate. Previously, the studio needed 90‑minute training sessions to onboard users. With Flow, they simply sent a link: “In many ways, agents are more natural to work with,” Pletten said.

Within two weeks, Flow had more active projects than ContentGo had accumulated in six months.

Planning for the future of AI

Looking to the future, Pletten outlined several concepts he believes will be crucial going forward. Some key things to keep in mind, he said, include:

1. Context engineering will become essential. Managing the limited context windows of LLMs and agents will remain a critical discipline. Systems like ContentGo and Flow are early examples of how organisations must design structures that help AI models understand and retain relevant information.
2. Everything will become an agent. Pletten said apps will disappear entirely: “Agents acting on our behalf don’t need an app layer or a graphical interface. Those are detours when AI can use APIs or command‑line tools instead.”
3. Multi‑agent loops will transform work. Agents already iterate internally, but Pletten expects loops to expand across multiple agents. For an agency like Partnerstudio, live performance data could eventually trigger new content iterations automatically.
4. Recursive self‑improvement is coming. Flow already includes a system that allows agents to learn from their mistakes. For now, humans remain in the loop, but Pletten said we are close to a “human above the loop” model, where agents improve themselves while humans supervise. He pointed to developments at OpenAI and Anthropic, where new coding models contributed to their own creation. If AI can soon conduct meaningful AI research, he said, predictions of reaching AGI — or artificial general intelligence — by 2027 or 2028 may be on target.

Despite rapid technological progress, adoption remains slow — especially in large enterprises.

“No one will be able to keep pace with the technological progress step for step,” Pletten observed. “But if you can stay ahead of your competitors, the upside is significant.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .