SeaDance 2.5 is presented as advancing AI video from face swaps to replacement of entire people or objects, including in 1990s VHS footage. The system is described as matching subtle details and handling subjects at different scales, partial occlusion and multiple depictions in one frame, although facial identity can drift and faces can jump.

Examples include a video of Dylan Patel as Drake, posted about a day earlier and described as convincing enough to fool many viewers, and footage showing Jordy transformed into Steve Jobs while retaining Jobs’ voice. An agent-enabled workflow using Codex could find source videos and reference photos without manual uploads. Video generation remains compute-intensive; blended production and possible use for reshoots are presented as likely applications.

Higgsfield wrapped CDance 2.5 in a harness designed to make the model easier to use. CDance 2.5 is available through an API, Fall, and several other interfaces, and the project is described as a fun, respectable effort even if completed by one person.

The model is characterized as highly compute-intensive. The assessment says China is “obviously way behind” on compute and raises, without resolving, what strategy fits those requirements.

Console builds AI agents that it says automate 70% of IT, HR, and finance support.

The agents are intended to give employees instant resolution for access requests and password resets.

Film productions are expected to adopt specialized production tools selectively, especially for urgent reshoots and possibly stunts, while continuing to rely on actors for most ordinary performances. Some of this use is already taking place in Hollywood for reshoots, according to the assessment presented.

Henry Cavill’s return for a Superman reshoot after growing a mustache for Mission: Impossible is cited as an example. The mustache was removed in post-production and his lip and mouth reconstructed; using 2017 technology, the result was described as very rough, tedious and expensive, and was widely criticized by the VFX community.

A video compilation uses real stunt footage featuring Johnny Knoxville, Steve-O and their crew while replacing the characters or suits with AI-generated elements. The underlying action remains real, making scenes such as stunts and aggressive motion look like AI prompts without generating the action from scratch.

The workflow was controlled by phone: Codex ran on a PC through the ChatGPT app, generated reference images with ChatGPT Images, and used them for character swaps through SeaDance, rendered via Higgsfield. The process required no direct coding. The assessment was that such demonstrations may spread as memes before becoming routine creative tools.

Higgsfield is criticized for maintaining separate API and plan-credit balances. A purchase of a plan reportedly left the credits unavailable through the API, while credits later added to the plan could not be used as API credits, leaving money distributed across different balances.

The service’s website is described as using countdown discounts and other tactics that encourage impulsive purchases. The account also references complaints about cancellation and unspecified allegations involving Higgsfield, without detailing or substantiating them. Despite the billing problems, the underlying product is described as effective for delivering video assets; trying Fall as a cleaner, more intelligible API is mentioned as a possibility.

AI work can be done properly, but it requires a discerning eye and taste.

Taste is described as the new moat in AI, with Fedora presented as part of that story.

An analysis argues that ByteDance may rationally prioritize AI video because it can connect model development directly to TikTok, Douyin, CapCut, advertising, e-commerce, and recommendation systems. That gives the company immediate distribution and potential benefits for its core business, without needing to create a new demand surface for video.

The analysis contrasts this with Anthropic and OpenAI, which are described as lacking an equally obvious distribution channel for video; OpenAI’s Sora attracted users but did not create the same flywheel. It also portrays Meta and Google/YouTube as more focused on agents, knowledge retrieval, coding, recursive self-improvement, and AGI. The analysis questions how China is producing strong video systems with reportedly one-fifth to one-third the compute, while noting that the models remain highly compute-intensive and that intellectual-property rules there may be relatively flexible.

An analysis frames training AI on transformative or fair-use video clips as a separate legal question from generating outputs that closely reproduce a person’s likeness or a copyrighted performance. Film criticism and edited clips posted to TikTok, YouTube, and Instagram may add commentary and not serve as substitutes for the original; such material could therefore enter training data. The analysis says this does not necessarily resolve liability for outputs that recreate someone such as Tom Cruise or reproduce a performance.

As an example, Google’s music-generation service is described as refusing an exact Madonna song while offering a generic pop song that, in theory, draws less from her work. Full movies uploaded to TikTok may be removed but could still pass through or be retained on ByteDance systems. The treatment of highly similar outputs remains uncertain and may be settled through future disputes and litigation.

ByteDance is not open-sourcing the discussed video model, its weights, or its training dataset—an approach described as diverging from the default Chinese AI strategy. The analysis argues that companies in the lead have less incentive to open-source models, while challengers may do so to build a developer ecosystem.

Open-weight coding models could support a Chinese technology stack and potentially help Huawei stay competitive, while video models have less incentive to be opened when a company is vertically integrated and controls distribution. The analysis contrasts this with early Meta, citing open models as a strategy for challengers that need an ecosystem.

AI video models may eventually become useful for sim-to-real translation and the simulated-video training of humanoid robots, potentially making them part of a broader robotics strategy. China also appears to be doing well in humanoid robotics.

This application is considered farther off than near-term entertainment uses; over the next month, attention is expected to remain focused on memes and funny videos.

A cited estimate says Meta rose 27% during the month, almost entirely because of Muse, though that causal link was questioned. Investors who tried the product reportedly gave it strong reviews, and Muse was described as temporarily leading a subcategory despite being built on a model that is not frontier-level.

Muse can be integrated into Meta’s shopping and messaging products; Shopify also announced a partnership with it. As Muse grew, Threads reportedly fell down the charts while Instagram prominently promoted Muse. The expectation expressed was that Muse could offer more monetization opportunities than Threads, including through Meta’s small-business base.

Amazon said it cut off Meta’s Muse personal AI agent from shopping on Amazon.com on customers’ behalf after unsuccessfully asking Meta to exclude the site. Amazon said it had not authorized Muse to access its store, that the agent did not identify itself while browsing, and that it appeared to capture and store customer credentials, creating privacy and security risks.

As of Sunday night, users trying to shop with Muse saw a message calling it an “unauthorized AI agent” and citing Amazon’s conditions of use. The dispute highlights the conflict between retailer-controlled shopping agents—such as Amazon’s Rufus—and independent agents that browse or transact for users.

Advertising inside AI agents could conflict with users’ expectation that the agent acts exclusively in their interest. A user objected when free service Instinct sent promotional invitations, viewing the outreach as benefiting the agent rather than the user despite the product’s operating costs.

The analysis considers sponsored agentic actions in ChatGPT and similar products as a possible future advertising format. Companies may pay for recommendations that create incremental demand, but a referral to Amazon is less valuable when the user already intended to shop there. Trust and perceived loyalty could be central constraints on monetizing agents.

Reliability concerns are resurfacing for AI agents: hallucinations in knowledge retrieval were perceived to have receded but are now described as returning. Frontier tasks such as system design, Blender and AI video remain more error-prone.

In one cited flight-booking case, an agent reportedly generated the middle name “Bom Chang Choi” for David, potentially disrupting boarding. The error may be corrected through airline support and a new boarding pass, but it could still delay security or cause a missed flight.

One explanation offered is cost pressure: free products served at scale may use older models or other cost-saving approaches. A cloud-hosted OpenClaw instance was tentatively identified as possibly running Kimi; that attribution remains uncertain.

Former Meta employee Nikita Beer says auto-negotiation was one of the launch features of the Muse app for Facebook Marketplace. He argues that millions of bots repeatedly lowballing sellers could eventually damage the product; the claim is presented as a concern, not an established outcome.

The discussion also considers bots sending outbound messages to potential buyers or sellers. If automation operates on both sides, Marketplace could see bot-to-bot negotiations or large-scale unsolicited outreach, though it remains uncertain whether that would offset or worsen the problem.

Rainmaker said it raised $100 million, describing the financing as small compared with the hundreds of millions of dollars China’s Meteorological Administration reportedly spends annually on weather modification. The company framed water as a strategic resource and a fundamental input to industry, agriculture, and food production.

Rainmaker identified cloud seeding as its foremost option for producing more water and argued that, without additional production, water bodies including the Colorado River and Great Salt Lake could become impaired to the point of mass evacuation. It said greater water production is needed for U.S. strategic health and for overseas allies such as Jordan and Morocco.

Rainmaker says it has closed one deal with a hyperscaler and is negotiating with multiple others on engineering feasibility for arrangements that would offset their water consumption by producing more water for the American interior.

The company’s statement comes amid debate over data-center water use: American agriculture and municipal consumption are orders of magnitude higher, but data centers still consume water, which Rainmaker says matters in the American West’s supply shortages.

Rainmaker says it is negotiating cloud-seeding deployments with ski resorts, including resorts in Switzerland, with activity expected to draw attention ahead of the upcoming winter.

The company attributes the interest to results it reports from Alaska, Oregon, Idaho and Utah, where it says it produced hundreds of millions of gallons of new water. Rainmaker is adding field sensors and building more sophisticated atmospheric models to measure interventions and water production in real time. The reported volumes and causal effects have not been independently validated in the supplied material.

Rainmaker says it is making a major strategic bet on autonomy, continuing to use its drones while making them increasingly autonomous. The company expects autonomy to enable more flights and cost-effective atmospheric observations, while lowering the cost per volume of water or snow delivered; it may also conduct research with planes in the not-too-distant future.

The business is seasonal and annually cyclical because of El Niño and La Niña. Rainmaker says it is geographically hedged across the United States and other countries, allowing it to shift work between regions with different snowfall and climate conditions, including areas facing depleted reservoirs.

Rainmaker says it wants to provide water services to the federal government and nations around the world.

To pursue that ambition, the company says it needs to collaborate with leading national laboratories and universities across the country. The company also credits its Gundo base as essential to building Rainmaker.

Rainmaker says the rate of hypothesis generation and experimentation at the company is accelerating because of available models.

It describes Astra as a “spectacular lever” and says it is developing better models with LLMs participating in the critique of outputs—an approach it considers genuinely additive to understanding the atmosphere and expects to become more useful over time.

Vinod Khosla expects personal agents to become a broad, diverse category rather than a winner-take-all market. He said major companies including Meta and Google will develop agents with different trade-offs, while startup agents may move faster.

Khosla said agents can handle browsing and research, present options, and ask users interactive questions. He identified advertising incentives and users’ willingness to trust companies with their data as fundamental differences between products. Alex Kerr argued that processes such as browsing and comparing options are part of the experience, and that personal agents may be absorbed into existing products rather than remain a standalone category.

Personal AI agents are framed as a potentially large category that represents users’ interests rather than those of a company, product or predetermined viewpoint. ChatGPT is described as already functioning like a browser: a single interface that can replace moving among 20 websites or Chrome tabs for research and product questions.

The consumer-aligned approach could conflict with platforms such as Amazon, whose existing businesses may be threatened by agents that mediate navigation and shopping. A personal AI could also protect users from marketing pressure and biased reviews, including reviews influenced by undisclosed promotion.

Personal agents may overcome the conventional assumption that consumers will not pay for productivity tools if they deliver direct savings. Proposed uses include canceling unused subscriptions, obtaining refunds, and finding cheaper options in shopping and travel.

ChatGPT’s subscription service is cited as evidence that consumers will pay when the value is high; a money-saving agent could be priced more like TurboTax than an entertainment product. It remains uncertain whether consumers would consistently pay around $20 a month, but the model is presented as an agent acting for the consumer rather than for a company.

Traditional data lock-in may be a weaker moat for personal agents because AI could automate the complex process of offboarding from one service and onboarding to another, although the process may initially be clumsy. The ability to move prompts and memories between OpenAI and Anthropic was cited as an existing example of this kind of switching.

The analysis identifies trust in a company and dependable task completion as potential primary moats. Meta was described as having a major disadvantage if consumers do not trust it with personal-agent data, while products that reliably get tasks done may retain users.

The Wajo agent can recognize when it is stuck and bring in a human to complete a user’s task, rather than simply reporting that it cannot proceed.

Human assistance is difficult to scale for very large companies: a product serving a billion users would need to build a substantial workforce. The discussion characterized being small and agile as an advantage, and suggested that users who need a task completed may prefer paying a little for human help over leaving it unfinished.

Personal agents may need to operate primarily through channels people already use, including email, text messaging, Slack and phone calls, rather than requiring a dedicated app or browser. Any app could remain in the background, while an agent might also call a merchant directly.

The proposed model emphasizes transparency: an agent should identify itself as an agent rather than impersonating the person it represents. In a Petco ordering interaction, the agent reportedly refused to pretend to be the user, with transparency framed as important for trust.

One shopping use case involved searching six websites for a specialized post-surgery dog collar and checking same-day delivery, including whether an agent could pay. Payments were presented as a potential differentiator among personal agents.

The Wajo agent was described as suitable for use with an Amex card because of certain protections intended to prevent unauthorized reuse; their details were not disclosed. Proposed safeguards include coded spending rules and thresholds, which could prevent a model error in which a purchase was interpreted as $100 instead of $100,000. Abuse, mistakes and security challenges remain risks, with future systems expected to combine deterministic software controls and stochastic AI.

Personal AI agents may eventually receive power-of-attorney-like authority to complete tasks that currently require users to intervene or call a service. An estimate cited in the discussion put potential adoption at about 5% of users today, rising to 40%–50% by 2030 as repeated interactions build trust.

Trust and transparency were identified as the key adoption advantage for agent companies, with startups that demonstrate strong transparency potentially gaining an edge over brands perceived as lacking trust.

Wadjo was backed by an investor on day one, after the team brainstormed the idea with Shivani, who was described as combining strong technical ability with an understanding of human factors and consumer needs.

The product is currently available only through limited, invite-only access. The team intends to scale gradually as Wadjo builds connectivity to more services, with a free release planned for September 30 and broader availability expected by the end of that month.

The FAA has issued a temporary flight restriction around SpaceX’s McGregor, Texas, facility ahead of Tesla’s Roadster reveal, referred to in the source as October 1. The restriction covers a 1.5-nautical-mile radius from the ground to 10,000 feet—an unusually high ceiling compared with restrictions previously associated with testing at the site.

The scope of the restriction is being treated as possible evidence of a demonstration involving the Roadster’s rumored SpaceX package, including cold-gas thrusters. Tesla has not confirmed that the car will fly or hover; alternatives include a remotely operated lift-off or an Evel Knievel-style ramp jump. One prediction put the height at about 10 feet, while other possibilities discussed ranged from 30 feet to much higher. The vehicle’s demonstration, production, safety and regulatory outcomes remain unknown.

The Tesla Roadster’s discussed commercial terms include a $45,000 deposit due within roughly 10 days, with the deposit described as potentially refundable if the reveal disappoints. A price near $250,000 was viewed as competitive for a supercar, while an $800,000 price would put the vehicle against stronger competition.

The original specification cited for the Roadster called for 0-to-60-mph acceleration in two seconds. A possible range of more than 600 miles was also cited, but the discussion said no EV had achieved that and expressed uncertainty about how it could be reached in a relatively small vehicle. These figures and the final price remain unconfirmed.

Box argues that AI agents will increase the importance of enterprise systems of record rather than reduce it. Large-scale deployments may involve thousands of end users and tens of thousands of agents, requiring coordinated access controls, business logic, user experiences, security and authoritative records—not just an underlying database.

The company says it has reaccelerated growth far beyond its internal plans as enterprises use core platforms to manage unstructured data, including contracts, research materials, marketing assets and financial documents. Box is positioning itself as a platform for storing, governing, retrieving, deploying and securing that data; it also says agents may shift value toward infrastructure and data platforms, although it does not specify which software segments will ultimately lose value.

Personal agents that use computers and browsers could move beyond research assistance to completing entire consumer tasks, potentially increasing demand for local services and marketplace activity. The analysis argues that reducing friction—such as arranging flower deliveries, family activities or installation help—could lead consumers to do more rather than simply save time; an agent was described as having booked a visit to San Francisco’s Japanese Tea Garden.

Secure authentication and a trusted handshake between agent systems and tools remain unresolved, while payments are described as largely solved. Broad consumer adoption is not considered certain: personal-agent companies could instead pivot to enterprise use cases and OAuth if the products fail to cross the adoption chasm. The analysis also suggests adoption may require limited behavioral change because agents could become a new way to use existing mobile apps.

A personal P(doom) estimate is described as “very, very low,” while extensive AI safety work—including cybersecurity-related work—is supported. The rationale is that sustained attention and incentives across the AI ecosystem could help identify and address catastrophic risks.

The same view warns that safety language may be co-opted by parties with different goals, creating a narrow line between funding safety and alignment research and generating political pressure to stop new data centers from being built. The AI policy debate is expected to remain messy for the coming years and become a dominant subject of the 2028 presidential election, with jobs, GDP and safety at issue. Conflicting industry positions are illustrated by support for both an open-weights initiative and a pacing letter.

iConnections describes itself as a capital-introduction platform connecting institutional investors with private-equity, venture-capital, private-credit and real-estate fund managers. It says the platform facilitates about 30,000 one-on-one meetings annually and moves approximately $10 billion to $20 billion through its ecosystem.

The business combines software that identifies relevant counterparties with events where meetings take place in person. Institutional investors attend for free, while general partners pay for the software-and-events package.

Started in 2020 with a global charity event benefiting food banks, iConnections says the event attracted about 400 institutional limited partners and generated 3,000 meetings. It says the disruption made the business more technology-driven, although face-to-face interaction remains important for large alternative-investment commitments.

Tomorrow X was conceived by Gavin Baker, Antonio Gracias and a third organizer as an effort to bring more technology executives into the private-markets community and broaden access to a world-class banking-conference experience.

The planned event is intended for retail investors and independent analysts active on Substack, Reddit and X who would not typically be invited to major Wall Street conferences. Organizers say it will feature prominent technology and finance executives, including Jensen Huang, Michael Dell, John Gray and David Solomon, as well as CEOs from Databricks and Vercel.

Rather than starting with a small hotel ballroom, the organizers rented a stadium and say the event will be priced inexpensively to make it accessible to a mass audience.

An analysis argues that AI safety debates often conflate several distinct issues, including data-center climate effects, bioweapons, and model sycophancy and its potential impact on people’s mental health. It distinguishes model-layer risks from risks arising where people interact with deployed AI applications.

The analysis says application-layer risks may receive insufficient attention from frontier labs, even though products used by Fortune 100 companies can reach millions of people and affect more. It also presents the view that doomer sentiment is concentrated at the model layer because AGI and superintelligence are expected to emerge there.

Raindrop, an agent-reliability product, recently announced simulations that model how changes to AI agents could behave in the real world and surface unexpected actions. The product is intended to detect production issues and help prevent them; companies define what they consider good or bad behavior rather than relying on a simple binary classifier.

The current approach replays selected historical events and generates on-the-fly adversarial scenarios for specific cases. Tool calls in past agent traces are used to reconstruct the surrounding simulated context, while the simulation is monitored for missing or inaccurate data that can be identified and filled at runtime.

As consumer agents become more popular, businesses may face more calls and messages from them and could deploy counteragents to process and negotiate with other agents. The analysis predicts a growing need for companies to adopt agents customized and fine-tuned for their operations, then test, evaluate, and improve how they perform.

Such alignment may reflect a company’s legal, operational, and cultural context rather than only the morality encoded by a model’s training lab. An off-the-shelf model may not be a cultural fit for a particular company, while companies may want software that helps them understand how their agents work.

Apple and Google are described as well positioned to build personal agents because users already entrust them with sensitive data. Apple is characterized as having especially high privacy trust, while Meta is seen as more tolerant of algorithmic errors.

The analysis argues that Apple’s design-led process—defining how a product should work and engineering toward that target—fits less well with AI’s uneven, stochastic capabilities. Frontier labs instead build products around what their models can currently do, such as strong Python performance alongside weaknesses in other areas. Apple is still described as “ridiculously well positioned,” and the analysis cautions against counting it out.

Fortune 100 companies are reported to show substantially less adoption of Chinese AI models, with perceived risk cited as a factor. Cost remains important as usage scales amid frontier-model price competition.

Companies are weighing when to use third-party coding agents such as Claude Code or Codex and when to build their own systems. The case for internal AI workflows is described as stronger than it was six months ago, partly because company-specific documents, ticket-routing paths and organizational structures are difficult to replicate in general-purpose tools.

The source also describes a low barrier to entry for agent-style products, making distribution important, and says Fortune 100 and Fortune 500 companies are increasingly building internal agents and applications that might have seemed unnecessary two years ago.

The analysis argues that as AI makes polished brand elements such as 3D assets widely accessible, companies need distinctive taste, cultural awareness and a clear sense that a real human team stands behind them. Brand can be understated and expressed through what a company does and its presence on Twitter, rather than only through its website.

It also argues that AI-generated writing is often recognizable through overused phrasing and formatting. Detectable text can feel disrespectful or oversaturated enough to make readers stop, even when its underlying quality may be comparable to ordinary writing; Codex responses were described as sometimes obnoxious.

A blog about the Hugging Face incident, accompanied by a simulated game called Hugging World, was labeled “vibe written” after AI assistance was used for some of the writing. The author said the blog was about 80% complete and contained useful information, but differed from one written sentence by sentence by the author; the exact AI-generated portions were not specified.

No pushback over the disclosure was reported.

A company raised $35 million from CRV.

The funding amount was corrected from an earlier mention of $25 million. CRV was identified as Charles Rivers Ventures, which was described as being in Boston.

Paramount said it reached a settlement with California and other states challenging its $111 billion acquisition of Warner Bros. Discovery, removing what had become the last major obstacle to the deal. The states had argued that the combination would give one entertainment company too much control over major studios, cable networks and other media properties.

The agreement calls for an independent board of journalists intended to protect editorial autonomy at CNN and CBS News, at least 30 theatrical releases per year—with 20 required to be wide releases—and commitments on domestic film and television production. Paramount will not sell CNN or divest a movie studio. The integration is expected to bring substantial job cuts, while it remains unclear which of the two companies’ high-profile streaming leaders will run the combined streaming business.

As the lawsuit moves toward resolution, prospects appear to have receded for converting part of Paramount’s Los Angeles lot into a large data center. The transcript does not establish whether a formal proposal had been advanced or the idea was only under discussion.

The possible data center was described as requiring substantial amounts of local water and energy; the property is instead expected to remain focused on movie production.

A claim that humans in India manipulate or answer ChatGPT responses was characterized as partly grounded in the human creation and curation of training data, but wrong about people manually answering each live query.

ChatGPT is described as generating responses at machine speed, while companies such as Scale AI and Merkur are cited as part of the broader data ecosystem. Professor Zhang was described as guaranteeing the claim, though the extent of human review was not detailed.

A speculative view portrays governments globally as potentially using podcast personalities who mix accurate and fabricated claims to attract attention and create confusion. The source provides no specific evidence that governments deliberately use such figures.

David Sacks is identified as an AI czar and co-chair of the President’s Council of Advisors on Science and Technology, abbreviated PCAST. The resemblance between “PCAST” and “podcast” is presented as a humorous observation, not evidence of an institutional strategy.

The current movie year is the highest-grossing since 2019 and has produced the second-highest number of billion-dollar movies on record, with major releases and YouTube-originated projects performing strongly.

However, ticket sales remain well below prior levels, partly because higher ticket prices are lifting grosses. Moviegoing has shifted from a routine weekly habit to an event-driven choice competing with TikTok, YouTube, live sports, concerts and other entertainment. The movie business is feeling better than it has since theaters closed during the pandemic, but it remains uncertain whether the improvement is a temporary blip or the start of sustained growth.

Netflix is increasingly signing major YouTube creators to reach younger viewers and increase viewing time. Newer arrangements can require creators to release videos on Netflix and YouTube simultaneously, with Netflix offering millions of dollars in addition to existing YouTube income and discussing possible original programming.

The reported response from YouTube includes reducing creators’ reach when they work with Netflix and rewarding them if they return. The competition may benefit creators in the short term by giving them more negotiating leverage and potentially increasing compensation; Disney, Fox and other legacy entertainment companies are also seeking creator deals.

Netflix could cultivate emerging creators by spreading funding across many smaller projects rather than backing only individual large productions. One proposal would give roughly 20–30 creators $50,000 each, helping the service build a distributed source of fast-turn, higher-quality content; however, it is unclear whether Netflix wants to compete in user-generated programming.

The analysis suggests Netflix could distinguish itself from YouTube through a curation layer—described as a potential “Slop Shield”—that filters videos made from AI scripts with little substantive content. This is presented as an opportunity, not an announced Netflix initiative.

Apple TV+ has built a portfolio described as well-reviewed and recognized with Emmy and Golden Globe awards, despite the hit-driven nature of film and television and the inevitable flops that can conflict with Apple’s culture of polished, “perfect” products.

Its standing as a Hollywood buyer remains uncertain. Apple TV+ appears to be sitting outside the Netflix–YouTube competition, but it is unclear whether Apple is satisfied with its current position or will expand and compete more directly for creator talent.

Streaming services are evaluating vertical dramas, including romantic, horror and true-crime formats, as growth slows and platforms seek more ways to attract viewers.

Peacock is described as having already added some vertical programming, while Netflix and other services are said to be examining it. Services are also looking at YouTube creators and podcasts, with the view that they will pursue attention-generating formats if they can integrate them sensibly; vertical dramas are described as less suited to Apple than to other platforms.

Netflix is facing questions about its future after what was described as a rocky year since it walked away from Warner Bros.

Discovery. Industry observers are watching for the company’s next growth plan and how it will define its next phase. There are also reported rumors of possible leadership changes, but whether they will happen remains uncertain.

Comcast is splitting its cable business from its entertainment business, a move compared with a restructuring Warner Bros.

Discovery had previously planned. The separation has prompted speculation that it could be a first step toward NBCUniversal being combined with another, unspecified entity.

AI is currently a bigger conversation in the music business than in Hollywood, according to the source’s assessment.

The issue is nevertheless described as an emerging concern in Hollywood and as something that is “lurking” across all sectors of society.

A push is underway for a federal tax credit for filming in the United States, with the stated aim of bringing production back from Europe and Canada. Donald Trump has expressed support for the proposal, and advocates are trying to advance it before the next Congress takes its seats in January.

It remains unclear whether Congress will approve the credit.