July 22, 2026

Media Transformation Summit — July 2026 — Recap

The third Media Transformation Summit brought together nine publishing leaders to examine what transformation looks like once the hype meets real organizations. Across strategy, production, formats, operations, data, metadata, and discovery, one conclusion kept returning: AI may lower technical and economic barriers, but it moves the difficult decisions elsewhere. Publishers still need judgment, rights clarity, quality controls, workable governance, structured data, audience understanding, and the discipline to learn before they scale.

Executive summary

Publishing transformation is not a race to install the most tools or automate the greatest number of tasks. It is the ability to make the next change easier and better.

The July 2026 Summit moved deliberately from organizational capability into production, formats, content systems, operational adoption, and reader discovery. Amy Jones established the central test: what will be easier next time? Fred Zimmerman demonstrated the radical production frontier created by agentic automation, while also showing that quality has to be engineered. Andrew Weinstein and Mohit Srivastava moved the bottleneck from production to rights, audience, channel, and portfolio judgment. Dmitry Shishkin showed why shared taxonomies and explicit user needs are becoming infrastructure. Arantza Larrauri brought the discussion into real organizations, where policies, incentives, architecture, and employee behavior determine whether experiments become capability or chaos. Sarah Arbuthnot, Emma House, and Thad McIlroy completed the picture by showing that technical discoverability is only the beginning. A book must still be understood, chosen, available, and connected with the right reader.

Taken together, the sessions did not produce one universal blueprint. They produced a more useful operating principle: choose a real problem, run a bounded experiment, measure what happened, preserve what is reusable, and make the next decision easier.

What the Summit made clear

Transformation is a capability, not a finish line. A project matters when it leaves behind reusable foundations, clearer decisions, and teams that can continue improving.
Cheaper production increases the importance of judgment. Rights, audience, format, timing, channel, and quality become more important when almost anything can technically be produced.
Automation quality must be engineered. Prompts are not a quality system. Publishers need deterministic checks, semantic review, explicit approval points, and clear accountability.
Structured content is becoming operating infrastructure. Shared taxonomies, user needs, precise formats, sourcing, and attribution connect editorial intent with product, analytics, and AI-mediated use.
AI adoption is an organizational design problem. Leadership, middle managers, employees, governance, architecture, incentives, and external partners have to move together.
Discoverability is not demand. Metadata, SEO, GEO, rich public content, positioning, availability, events, and audience insight work cumulatively.
Small experiments are more useful than broad declarations. The most credible next step is often a focused test designed to reduce one important uncertainty.

Evidence highlights

Emerald built a reusable service layer between a transformed front end and legacy systems.
Emerald identified 65 possible improvements in one customer-service area and implemented 21 in less than a couple of months.
Fred Zimmerman reported publishing 322 books in six months and described a pipeline with 108 deterministic file checks.
Andrew Weinstein described Scribd using subscriber behavior and available rights to identify backlist audiobook opportunities.
Mohit Srivastava proposed a five-episode microdrama test costing less than roughly $100 before funding a full adaptation.
Dmitry Shishkin showed an example where 80% of output generated only 8% of impact.
Arantza Larrauri described how a useful internal tool was stopped because no architecture or approval route had been agreed.
Sarah Arbuthnot cited Nielsen BookData findings linking complete and early metadata with materially higher sales.
Emma House described a campaign that found younger and more female buyers when values and emotions replaced demographic assumptions.

A practical 90-day plan

Choose one real business problem rather than the goal of "adopting AI."
Name an executive sponsor, an operational owner, and the people responsible for quality, rights, security, and customer impact.
Publish a minimum AI and automation policy covering approved tools, confidential material, review, disclosure, and external partners.
Map the current workflow before adding technology.
Audit the rights and commercial position of a focused backlist sample.
Review metadata completeness and consistency for a commercially important set of titles.
Enrich a small number of public book pages with original, authoritative information.
Define a minimum shared taxonomy for the pilot: topic, user need, precise format, audience, platform, and intended outcome.
Run one bounded format, workflow, or discovery experiment.
Measure behavior and operational results, then document whether to stop, adapt, or scale.
Capture reusable prompts, checks, decision rules, integrations, training, and lessons.

Important tensions

Automation versus craft: AI can perform more production work, while the appropriate level of human authorship, review, and judgment remains contested.
Volume versus reader scarcity: Content supply can grow much faster than attention and readership.
Dynamic metadata versus fundamentals first: Selective updating can reveal opportunity, but not at the expense of complete core metadata.
Openness versus control: Publishers want AI systems to understand their books while remaining concerned about licensing, training data, and unauthorized use.
Central governance versus bottom-up experimentation: Employees will experiment, but unmanaged experimentation creates security, quality, and integration risks.
Internal capability versus specialist partners: New tools make more work possible internally, but building every capability may not be the best use of publisher capital and attention.

Leadership questions

What will genuinely be easier the next time we make a comparable change?
Which workflow problems are people already solving through spreadsheets, personal accounts, or unofficial workarounds?
Where are we rewarding output speed while discouraging verification and judgment?
Which rights do we control but fail to exploit?
Which titles show evidence of unmet demand or an unexpected audience?
Does our website add authoritative information, or merely reproduce a retailer feed?
Can editorial, marketing, product, metadata, and technology describe the same content in the same terms?
What information do AI systems need to retrieve, verify, attribute, and recommend our content without flattening its meaning?
What is the smallest experiment that would materially reduce uncertainty within 90 days?
What reusable capability will remain after the pilot?

The bottom line

AI is lowering some of the technical and economic barriers in publishing, but it is not removing the need for judgment. It is moving the real bottlenecks toward rights, quality, workflow design, governance, structured data, audience understanding, channel choice, and availability.

The publishers most likely to benefit will not be those attempting to transform everything at once. They will be those that choose focused problems, run bounded experiments, measure honestly, and preserve what they learn. The most useful measure of progress is not how many tools were introduced or how much content was produced. It is whether the organization became better at making the next decision, running the next experiment, and adapting to what actually happened.

Sessions

Back to July 2026