The Weekly Bulletin | December 16, 2025

In partnership with

Hi SODP community,

We're two weeks from the end of 2025, a year that reshaped publisher economics in profound ways. Let's dive into what matters as we close out the year.

.TIP OF THE WEEK.

Tips on how to leverage Agentic AI for 2026

As we look toward 2026, we're witnessing a pivotal moment in enterprise AI adoption. Agentic AI systems that can act autonomously to achieve goals is transitioning from experimental pilots to production-scale deployments. This shift represents more than just technological maturation; it's a fundamental change in how organisations operate.

Here's what leaders need to know about this transformation and how to prepare for scaled agentic AI deployment:

1) Infrastructure Readiness: The Foundation for Scale

Moving agentic AI from pilot to production requires robust infrastructure that most organizations don't yet have. Unlike traditional AI models that simply predict or classify, agentic systems need to interact with multiple tools, databases, and APIs while maintaining security and compliance.

👉 Actionable Tip: Audit your current infrastructure for API connectivity, authentication systems, and observability tools. Invest in orchestration platforms that can manage multi-step agent workflows and provide proper logging for compliance and debugging.

2) Trust and Governance Frameworks Become Critical

In pilot phases, companies can tolerate occasional errors. At scale, agentic AI systems making autonomous decisions require comprehensive governance frameworks. Organizations must define clear boundaries for agent autonomy, establish approval thresholds, and create rollback mechanisms.

👉 Actionable Tip: Develop a tiered autonomy model where agents have different permission levels based on risk. Implement human-in-the-loop checkpoints for high-stakes decisions and create transparent audit trails for all agent actions.

3) From Single-Purpose to Multi-Agent Orchestration

Early pilots typically focus on single-use cases—a customer service agent or a data analysis assistant. Production-scale deployment means coordinating multiple specialized agents that work together, each handling distinct tasks while sharing context and goals.

👉 Actionable Tip: Design your agentic systems with modularity in mind. Create specialized agents for distinct functions (research, analysis, execution) that can collaborate through standardized protocols. This approach enables easier scaling and troubleshooting.

4) ROI Measurement Shifts from Proof-of-Concept to Business Impact

Pilot success is often measured by technical feasibility—"Does it work?" Production requires demonstrable business value: cost savings, revenue generation, or productivity gains that justify the investment.

👉 Actionable Tip: Establish clear KPIs before scaling. Track not just agent performance metrics (accuracy, response time) but business outcomes (customer satisfaction scores, operational cost reduction, time-to-resolution). Build attribution models that connect agent actions to business results.

5) Data Quality and Access Become the Bottleneck

Agentic AI systems are only as effective as the data they can access. At scale, data fragmentation, inconsistent formats, and access restrictions become major obstacles. Organizations discover that their biggest challenge isn't the AI, it's the data infrastructure.

👉 Actionable Tip: Prioritize data integration and standardization efforts. Implement semantic layers that give agents consistent access to enterprise data regardless of source. Consider investing in knowledge graph technologies that help agents understand relationships between data points.

6) Change Management: Preparing Teams for AI Collaboration

The shift to production-scale agentic AI fundamentally changes how teams work. Employees need to learn when to delegate to agents, how to verify agent outputs, and how to collaborate effectively with autonomous systems.

👉  Actionable Tip: Develop comprehensive training programs that go beyond tool usage to teach AI collaboration skills. Create clear escalation paths when agents encounter problems, and establish feedback loops where human users can improve agent performance over time.

7) Cost Management and Resource Optimization

Pilots run on limited budgets with predictable costs. Production-scale agentic AI can consume significant computational resources, especially when agents make multiple API calls or process large datasets autonomously. Runaway costs become a real risk.

👉  Actionable Tip: Implement cost guardrails and resource quotas for agent operations. Use caching strategies to reduce redundant API calls, and establish monitoring systems that alert teams to unusual resource consumption patterns before they impact budgets.

Organizations that succeed in scaling agentic AI will be those who:

• Build a robust infrastructure with proper security and observability

• Establish clear governance frameworks with appropriate autonomy boundaries

• Design for multi-agent collaboration rather than siloed solutions

• Focus on measurable business outcomes, not just technical capabilities

• Prioritize data quality and access as foundational requirements

• Invest in change management and team readiness

• Implement proactive cost management and resource optimization

The transition from pilot to production isn't just about technical scaling; it's about organizational readiness. Companies that approach this shift strategically, with attention to infrastructure, governance, and human factors, will unlock the transformative potential of agentic AI.

The question isn't whether agentic AI will move to production at scale; it's whether your organization will be ready when it does.

Shoppers are adding to cart for the holidays

Over the next year, Roku predicts that 100% of the streaming audience will see ads. For growth marketers in 2026, CTV will remain an important “safe space” as AI creates widespread disruption in the search and social channels. Plus, easier access to self-serve CTV ad buying tools and targeting options will lead to a surge in locally-targeted streaming campaigns.

Read our guide to find out why growth marketers should make sure CTV is part of their 2026 media mix.

Learn more.

.NEWS OF THE WEEK.

➡️ Google's AI partnership deals repeat Showcase playbook with superficial commercial agreements masking content access grab. Announced pilot AI partnership program with national news organizations mirrors Showcase framework: peripheral or non-products with real intention securing ongoing content access for core products (AI use) and heading off legal/regulatory challenges. Google pays "for extended display rights and content delivery methods like APIs" not for AI use itself. AI Overviews suppress clickthrough rates 35-50%, AI Mode results in only 5% of searches clicking to destination websites. UK Competition and Markets Authority designated Google "strategic market status" for search, EU opened investigation into AI Overviews, US Penske Media suing Google. Google has two options: create genuine marketplace for content with real licensing deals paying publishers on usage, or repeat Showcase playbook ensuring content access on Google's terms. Deals strongly suggest preferred route remains second option.

READ MORE

➡️ WSJ visual storytelling team breaks design rules to test audience behavior, collecting data that informs future projects. Hired team tasked with deviating from traditional styles, color palettes, typography. Designers prioritize projects with specific audience focus or experimentation goals. Recent experiments: in-article audio engagement (users more likely interact based on content than position), quiz length optimization (five questions achieved 70%+ completion), reaching younger audiences through auto-playing vertical video (over-indexed with younger/female readers). Each experiment provides actionable insights into reader behavior. Keeping decision-making process open ensures judicious project selection through form-driven Slack-based system. To break rules, you need rules to break.

READ MORE

➡️ Fragmented audience data across multiple platforms costs publishers growth through lost time, missed precision, slow product iteration. Omeda report shows 85% of publishers say audience data is competitive advantage, yet only 9% rate approach as very effective. Gap between knowing and doing: 64% analyze data, 57% have strategy, but only 45% update strategy based on insights, just 36% use data to personalize. Publishers manage 7-12 disconnected systems creating "human API work," manually exporting lists, cross-referencing spreadsheets, stitching profiles. Teams spend dozens of weekly hours reconciling lists instead of building products. Fragmentation duplicates functionality across 5+ tools, burns audiences with spray-and-pray marketing, slows product launches. Consolidation unlocks lifetime value tracking, smarter bundling, better targeting. AI, personalization, premium CPMs, subscriber value all depend on precision requiring consolidation.

READ MORE

➡️ Nieman Lab 2026 journalism predictions explore AI integration, business model shifts, and platform dynamics reshaping industry. Annual collection gathers perspectives from digital media leaders on coming year. Predictions cover emerging AI tools transforming newsroom workflows, subscription model evolution as advertising continues decline, social platform changes affecting traffic patterns, local news sustainability experiments, trust-building strategies in misinformation era, regulatory developments around tech platform accountability. Provides comprehensive view of challenges and opportunities publishers will navigate in 2026.

READ MORE

➡️ Publisher AI monetization strategies fail to replace advertising losses at scale, analysis shows. Publishers' ad share dropped 71% to 27.2% over decade, zero-click searches jumped 56% to 69% in 12 months. AI Overviews create 80-83% zero-click rates. Mid-sized publisher (10M monthly visits, $3 CPM) loses $168K monthly. Proposed alternatives insufficient: AI licensing $816.7M globally (2024) equals just 1.5% of ad blocking losses ($54B same year). Subscriptions scale from tiny bases, commerce content suffered 50% affiliate revenue drops post-AI Overviews. Best-case alternatives (subscriptions, licensing, podcasts, events) might generate $1.5M annually against $2M+ losses. Broken value exchange: AI trains on content, answers without sending users, publishers capture zero value. What would work: content marketplace for AI applications with real-time APIs, automatic compensation per inference. Infrastructure doesn't exist yet.

READ MORE

.JOB BOARD.

➡️ The Ferris-UC Berkeley Psychedelic Journalism Fellowship is open to journalists creating audio and print stories on psychedelics, offering a $10,000 grant to support in-depth reporting on this emerging beat. (Fellowship).

SEE MORE

➡️ The Global Investigative Journalism Network needs pitches on tips, tools, innovation, and strategies used in investigative journalism, seeking practical guidance that helps investigative reporters improve their craft. (Freelance).

SEE MORE

➡️ Defector needs summer editorial interns working 40 hours per week at $25/hour, totaling $10,000 for the 10-week program supporting independent sports and culture journalism. (Internship, Summer).

SEE MORE

.SOCIAL MEDIA.

➡️ Craig Swerdloff on LinkedIn:

"The reason newsletter sponsorships underperform (It's not what you think)."

Swerdloff, listening to Jeanne DeWitt Grosser on Lenny Rachitsky's Podcast, identified the core tension in newsletter sponsorship ecosystem. Publishers think their audience is worth more. Performance and Growth Marketers expect better performance. Both sides assume the other is the issue. Both sides are right.

The root cause isn't a performance gap, it's a measurement gap.

Most marketing teams aren't tracking the full customer journey or funnel. They judge sponsorships solely on "last click" conversions. Because upper and middle-funnel impact isn't measured, marketers underinvest in channels that build demand, and publishers lose renewals.

If we fix measurement, we fix both sides of the market.

Swerdloff's operational response at Wellput: redesigning onboarding process to focus on customer journey instrumentation, moving beyond "last click" to justify renewals, deploying AI agents to handle lead qualification so humans focus on high-value strategy.

For publishers, this diagnosis matters critically. Newsletter sponsorship underperformance isn't about audience quality or creative execution, it's about attribution models that systematically undervalue top-of-funnel awareness and consideration. When marketers optimize solely for last-click conversions, they miss newsletter sponsorships' actual contribution: introducing brands to qualified audiences, building familiarity, priming consideration that converts weeks or months later through other channels.

The solution requires educating sponsor prospects on proper measurement frameworks. Publishers should insist on multi-touch attribution, brand lift studies, or at minimum extended attribution windows that capture delayed conversions. Without measurement infrastructure that tracks full journey, newsletter sponsorships will continue appearing to underperform despite driving meaningful business impact that's simply invisible to last-click models.

CONTINUE READING

These are the highlights for the last week.

Until next!

Vahe Arabian and the editorial team at SODP