Inside ad tech before Cannes: agents, AI search, streaming, and trust

The week and a half leading into Cannes Lions 2026 produced an unusually dense run of news, and a single thread runs through almost all of it. The advertising industry is committing to automated, agent-driven decision-making at speed, while the measurement, verification and legal scaffolding that would make that automation trustworthy is still being assembled. The macro numbers set the stakes. WPP Media's midyear forecast, published June 18, projected 4.4 percent global advertising growth to 1.3 trillion dollars in 2026, crediting AI investment with offsetting geopolitical headwinds across every major channel. What follows is a guide to how that money is being reshaped, organized by the forces doing the reshaping.

The reporting period itself is instructive. It spans the days when product teams ship their most ambitious work, timed to the industry's largest gathering, and so it captures the distance between what is announced and what is operational with unusual clarity. The launches are real, the demonstrations work, and the spending intentions are documented. What remains unsettled is whether the data feeding these systems is clean, whether their decisions can be inspected, and whether the rules governing them will arrive before or after the budgets do. Read in sequence, the announcements describe an industry building the engine and the road at the same time, with the engine running ahead. The sections that follow trace that imbalance across the agent layer, the search surfaces, streaming, retail media, measurement, the platforms and the wider economy, because the same tension recurs in each.

The agentic buying layer goes public

The clearest signal of the moment is the rush to build the plumbing for machine-to-machine ad buying, and Cannes Lions became the stage for it. Microsoft used the festival to introduce Web IQ grounding APIs, citation reporting inside its free Clarity analytics product, and a Model Context Protocol server built to connect advertisers to live AI workflows. Hours of coverage earlier, a cluster of NVIDIA's ad tech partners, among them Criteo, Taboola, Alembic and KERV, detailed GPU-powered infrastructure for autonomous bidding, creative production and measurement. Pinterest arrived with its own MCP server, a Business Assistant, a new Performance+ model and an experimental consumer app called Ask Pinterest. Three companies, one component built under different names. MCP, the emerging standard that lets a large language model call external tools and data in a structured way, is becoming the connector everyone wants to own, because the platform hosting the most useful server becomes the place agents reach first.

The framing that ties these launches together was set out in a single analysis: agentic ad tech is trying to take over the buying layer as AI search budgets surge, with WPP projecting AI search reaching 39 percent of search revenue by 2031 and Adobe finding that 86 percent of consumers make impulse purchases monthly. The sell side moved in step. PubMatic opened what it bills as the first programmatic connected-TV auction for independent creators, routing premium inventory to agentic buyers through its AgenticOS framework, while LiveRamp opened its agent network to outside builders through a new LAB program spanning planning, activation, measurement and data transformation.

Verification and data vendors are racing to keep up. DoubleVerify launched DV Neura, a cognitive AI engine with MCP integration, agentic execution agents and what the company describes as a nearly 300-fold rise in content classification throughput. Databricks brought its CustomerLake agentic CDP to market with IAS as a launch partner, linking more than 300 billion daily media signals to first-party audience profiles. DeepIntent launched Helix AI, positioned as the first agentic planning platform for healthcare marketers, claiming to halve planning cycles with HIPAA-compliant natural language tools. On the supply side, Mediavine and SWYM expanded their partnership on June 16 to run AI-native bid-stream optimization across 18,000 publisher sites before any bid is placed.

Investment intent points the same way, even as deployment lags. Mediaocean's H2 2026 report named AI media the fastest-growing category at 60 percent, while flagging data gaps and stack integration as the obstacles slowing real adoption. The mismatch between ambition and readiness is the recurring caveat. Optable's agent readiness self-assessmentmapped six pillars publishers must fix before agentic buying reshapes how inventory is discovered and transacted, and concluded that most are not prepared. IAB Australia's 2026 agentic AI search report described a funnel that collapses into seconds and argued that trust becomes the scarce currency once an agent, rather than a person, makes the shortlist. That shortlist is itself becoming contested ground: Koddi research found 84 percent of commerce media leaders would invest in AI recommendation visibility, with 70 percent already using agents for campaign execution.

The structural questions sit underneath the product news. HubSpot is projecting 42 billion dollars in partner revenue by 2030, per IDC, with AI-first revenue growing at a 28.4 percent compound annual rate as agents move into the CRM and mid-market software stack. Independent agencies see opportunity in the shift: Brainlabs Chief Growth Officer Ali Reed argued that smaller shops are seizing the high-margin, AI-driven layer of work that legacy holding companies are ceding. Skepticism is warranted too. Perplexity co-authored a study reporting that AI agents cut task time by 87 percent, using its own data and products to do so, even as copyright and security cases mount against the company. And the traffic itself is not yet a straight line up: HUMAN Security's May 2026 data showed agentic traffic falling 4.3 percent month over month while blocking rates climbed toward 9 percent, as Comet, Atlas and Claude competed for the same requests.

Two observations cut across the agentic launches. The first concerns concentration. A shared protocol lowers the cost of connecting any agent to any platform, which in principle widens competition and lets smaller buyers reach inventory that once required custom engineering. The same standard also rewards whoever hosts the richest server, since an agent reaches first for the place where the data and the available actions are deepest, and open protocols have a long history of producing concentrated outcomes. The second concerns accountability. When a human buyer optimizes a campaign, the reasoning is at least partly legible in the choices made and the notes left behind. When software executes thousands of micro-decisions a second across bidding, creative and measurement, the audit trail becomes a technical artifact that few buyers are equipped to read. The vendors answering the integration problem at Cannes have not yet answered the inspection problem, and the distance between the two is where the real risk sits. For the agencies whose value once rested on manual optimization, the contest relocates rather than disappears, moving from who pulls the levers to who sets the objectives, supplies the data and verifies the result after the fact.

The technical detail matters here, because MCP is not a product but a convention, a way of describing tools and data so that any compliant model can call them. That neutrality is its appeal and the source of its competitive stakes, since the convention favors whoever exposes the most useful endpoints rather than whoever owns the model, and it is why a search engine, a social platform, a verification vendor and a supply-side platform all shipped servers in the same week.

AI is rewriting discovery, and breaking what measures it

If agents assume clean inputs, the reporting on AI-mediated discovery suggests the inputs are anything but. Lily Ray documented a feedback loop she calls AI slop, in which systems ingest fabricated search-optimization claims, repeat them with confidence, and seed the next round of content and training data with the same errors, a cycle that resists correction once a falsehood is embedded. The quality problem is measurable. Gracenote tested 2,600 titles in 13 countries and found that ungrounded large language models hallucinated all metadata for nearly one in five titles, with United States cast accuracy at just 53 percent, a direct argument for the grounding layers that vendors like Microsoft are now selling.

Measurement is the second casualty. Similarweb traced AI recommendations to real traffic and found AI-recommended brands 2.5 times more likely to receive a site visit within seven days, yet 56 percent of that traffic arrives through branded search rather than a direct click from the AI interface, leaving the influence undercounted in last-click reporting. The lesson about where attention actually forms is reinforced by Fractl, working with SparkToro data, which examined 358 data points across eight industries and found niche publishers deliver 1.7 times the audience affinity of high-traffic outlets. Content strategy is shifting accordingly: NP Digital surveyed 500 marketers in May 2026 and ranked original research first for AI search citations at 82 percent, with video content last at 2 percent.

The human cost of these shifts is concrete, and it falls on independent publishers. Google AI Overviews drained the search traffic of two publishers and pushed one site to close after 21 years. One of those casualties was overfishing.org, a 21-year-old independent resource that shut down once AI Overviews erased its referral traffic, and it was not alone: All About Berlin lost 70 percent of its search traffic to the same feature, threatening eight years of independent publishing on Berlin immigration. The macro picture from Oxford's RISJ 2026 Digital News Report confirms the trend at scale: social media now beats publisher websites as a news source in 30 of 48 markets, and AI chatbot news use has climbed to 10 percent from 7 percent a year earlier.

Bots compound the strain on infrastructure. Kinsta analyzed 10 billion requests and found AI crawlers trapped in query-string loops, hammering WooCommerce cart and checkout pages as many as 3.75 million times in a single day. Authenticity tools are emerging in response. Deezer launched a free AI music detector in 27 languages that scans playlists across 20 streaming platforms, reporting that 43 percent of switchers carry synthetic tracks. Even the most prominent figures are registering discontent with the surfaces where most search money is spent: Paul Graham, co-founder of Y Combinator, criticized the way sponsored placements appear inside image search results on June 15 and said he had switched to Bing.

The common subject across these reports is trust in the channels that translate intent into action, and the agentic infrastructure being built on top of them assumes those channels are clean enough to automate. The evidence suggests otherwise. Fabricated advice circulates inside AI answers, genuine AI-driven demand hides inside branded search, hallucinated metadata corrupts catalogs, and bot traffic distorts the logs that everything else is measured against. Automating a noisy channel does not remove the noise; it moves the noise upstream, into systems that act on it without a person pausing to ask whether the input was sound. The publisher casualties give the abstraction a human shape. A site that took two decades to build can lose the traffic that sustains it within months of a single product change at a search engine, and the model that replaces it sends a far smaller share of visits back to the source. The phrase trust becomes the currency, used in the agentic search research, reads less as a slogan than as a description of the only durable asset left when discovery is mediated by machines that decide what to surface and what to omit.

There is a measurement irony in all of this. The systems most capable of generating demand are the least visible in the reports used to allocate budget, so the channel doing the work goes unrewarded while the channel receiving the click takes the credit, and correcting the distortion requires instrumentation most stacks do not yet possess.

Regulators move on AI, privacy, and minors

Policy is catching up to the technology on several fronts at once. Anthropic updated its privacy policy on June 8, effective July 8, adding biometric identity checks, third-party KYC data and disclosures covering agentic data flows through Claude accounts, a sign of how identity verification is migrating into AI products. Vermont enacted two laws in June that ban AI-only mental health services and raise data broker annual registration fees from 100 to 900 dollars, with expanded disclosure requirements attached. The boundary between AI assistants and sensitive personal data is being tested commercially as well: OpenAI launched a ChatGPT personal finance preview for United States Pro users, pulling bank data through Plaid, applying GPT-5.5 Thinking reasoning, and promising 30-day deletion on disconnect.

Europe moved on age verification. The Court of Justice of the European Union, in case C-188/24, defined when member states can legally require age verification on pornographic websites established in other countries, a ruling with implications for any platform serving age-restricted content across borders. Britain went further on minors and social platforms, with the UK government announcing a ban on social media for under-16s, backed by regulations slated for Spring 2027 covering Snapchat, TikTok, Instagram, YouTube and others.

Publishers are taking the fight to AI scrapers through novel legal mechanics. UK publishers launched Search-Only Contracts to bill OpenAI and Google 500 pounds per scraped article, using county courts to enforce payment without the cost of intellectual property litigation. The United States is legislating disclosure: New York's Assembly passed bill A11292 on June 5, requiring AI crawlers to disclose their identity and purpose to news publishers or face penalties of 15,000 dollars per day. Taken together, the regulatory cluster shows lawmakers and publishers converging on the same target, the unconsented use of data and content to feed AI systems, from opposite directions.

The deeper pattern is that identity and consent are being rewritten at the infrastructure level rather than the policy level alone. Biometric checks and third-party verification entering an AI assistant change what it means to hold an account, well before any regulator weighs in. Age-verification mandates push platforms toward collecting more sensitive data to prove compliance, a tension between protecting minors and minimizing data that has no clean resolution. The scraper disputes, meanwhile, reframe a copyright question as a billing question, sidestepping years of litigation by turning unconsented access into an invoice enforceable in small-claims venues. Each of these moves shifts the cost of compliance onto a different party, and the advertising systems sitting downstream inherit whatever data, restrictions and liabilities result.

What gives the cluster its weight is timing. These measures land as agentic systems are being designed to move data across services automatically, which means the rules being written now will shape what those agents are permitted to read, retain and pass along. A privacy regime built for a world of forms and cookies is being asked to govern a world of autonomous data flows, and the gap between the two is where the next round of disputes will form.

Streaming and connected TV expand on every front

Connected television produced the densest cluster of product and deal news, and the headline transaction reset the board. Fox Corporation agreed to acquire Roku for 22 billion dollars, joining live sports, news, Tubi and The Roku Channel into a single CTV advertising and streaming platform, a deal that also anchored a wider reshaping of the programmatic stackin which Publicis and The Trade Desk ended their public dispute and UK publishers pressed their scraper claims.

Format and platform innovation followed. Teads launched CTV Ensemble on June 18, folding HomeScreen placements, in-stream video, dynamic creative optimization and AI-driven lower-funnel performance tools into one global platform, a bid to make connected TV answer for outcomes rather than reach alone. Philo rolled out pause ads across all of its ad-supported channels, available programmatically through Magnite, Index Exchange, Kargo, Nexxen and additional supply-side partners. The format has data behind it: WunderKIND Ads released the first CTV pause ad benchmarks, drawing on millions of programmatic impressions across 15 verticals with TVision attention data, and found pause formats roughly doubling the measured attention of conventional spots.

Measurement crossed into new territory. EDO brought its outcomes measurement to Roblox, benchmarking branded events and video campaigns on the gaming platform against convergent television norms, an effort to make game-world inventory legible to planners who think in reach and impact. Visibility into the open market improved as well: Pixalate's OpenEPG Index ranked 5,108 streaming shows by open programmatic CTV ad spend and consumer reach across 210 United States markets. Workflow tooling is converging with social-style buying: Smartly integrated with the Roku Ads API, letting performance marketers launch, optimize and measure CTV campaigns using familiar social workflows.

The structural deals reorganized who controls inventory. Netflix and TF1 agreed to place the French broadcaster's TF1+ programs, live broadcasts and major French sporting events directly inside Netflix, with TF1 retaining full control of its advertising inventory, a partnership model rather than a turf war. In Germany, Zattoo named Stroer its exclusive ad sales partner for CTV and video inventory across smart TVs, streaming players and mobile devices. Local television is automating too: Gray Media, the largest United States local TV station owner, partnered with Madhive's AI DSP and its Maverick platform to target 117 markets and the 2026 midterm advertising cycle.

The case for ad-supported streaming keeps strengthening. VAB research reported that 89 percent of United States streamers use ad-supported services, that FAST channels now reach 1,700 titles, and that interactive CTV ads produce 138 percent higher brand recall. A separate VAB addressable TV guide found 92 percent of pay TV households are addressable-enabled, with nearly half of current advertisers raising budgets this year. Audience composition matters in the planning: VAB data on Black audiences showed Black adults spending 44 hours weekly on TV and streaming, 38 percent more than the general population, with higher ad engagement across retail, food and fashion. Live sports remains the magnet for regulated budgets: DeepIntent launched verified live CTV inventory with custom pacing and pharma KPI measurement for healthcare brands targeting World Cup, NFL and NBA audiences.

The logic binding these announcements is consolidation paired with fragmentation, two forces pulling at once. The Fox-Roku transaction concentrates content, distribution and an advertising platform under one owner, the kind of vertical integration that promises efficiency and raises questions about who sets the price of inventory. At the same time, every new format, marketplace and integration adds a surface that buyers must learn, measure and reconcile against the rest of their spending, so the channel grows more capable and more scattered in the same motion. The attention case for connected TV is strengthening on the evidence, from pause-ad benchmarks to interactive recall figures, yet attention measured on one platform rarely translates cleanly to another, and the planner stitching a campaign across HomeScreen placements, pause inventory, FAST channels and live sports is reconciling metrics that were never designed to agree. The distribution deals in France and Germany suggest a workable answer, in which streamers and broadcasters act as partners rather than rivals for the same budget, but whether that model survives in markets where the two guard their inventory more jealously is the open question the period leaves behind.

Local and regional television sit inside the same dynamic. The move by the largest United States station group to an AI-driven demand-side platform for the midterm cycle shows national-grade automation reaching markets that political money floods every two years, while the German sales partnership shows broadcasters outsourcing the commercial layer rather than building it, two routes to the same programmatic destination.

Retail media tightens, and Amazon shows both edges

Retail media spent the period proving both its value and its limits, with Amazon at the center. Amazon Publisher Cloud began routing commerce signals into FreeWheel streaming TV deals, letting sellers optimize inventory against Amazon's purchase and shopping data, with early tests showing a 33 percent lift in on-target reach and Warner Bros. Discovery and A+E among the first participants. The same machine showed strain at its core: CommerceIQ data found Amazon brands entering Prime Day 2026 with higher return on ad spend and stronger inventory yet stockout losses up 24 percent, concentrated in the highest-selling ASINs, a reminder that a retail media impression collapses in value when the product behind it runs dry.

The sector's measurement gap is well quantified. Incremental's meta-analysis of 150,000 campaigns and 350 million dollars in ad spend found siloed retail media attribution misses 36 percent to 53 percent of true campaign ROI. That gap helps explain the demand for the impulse-buying behavior advertisers are chasing: Adobe surveyed 1,003 United States consumers and found 86 percent make unplanned online purchases monthly, driven by social video, flash sales and Gen Z stress-relief patterns.

Amazon also expanded the controls and surfaces available to advertisers. Amazon added audio as a supported inventory type in its DSP Campaign Management APIs on June 11, joining display, streaming TV and online video, and it added an Off-Amazon ad serving column to Sponsored Products bulksheets on June 8, letting United States advertisers control off-site spend in bulk. Catalog hygiene is changing too: Amazon will cap product titles at 75 characters on July 27, with implications for Item Highlights, AI rewrites and Prime Day timing.

Commerce media is spreading beyond the marketplace giants. Koddi connected its retailer ecosystem to Search Ads 360, letting advertisers run and measure commerce media campaigns inside existing SA360 workflows, while Microsoft Advertising launched Product explorer for United States advertisers with under 100,000 SKUs, offering a unified view of product status, serving and performance. The delivery platforms are becoming media networks in their own right: Uber Advertising launched Offers on Uber, new Uber Eats premium formats and Offsite Ads to extend brand reach beyond its own ecosystem, and followed with Deal Drops and Reorder Rewards on June 8, giving restaurants time-bound event offers and post-purchase repeat mechanics.

Retail media's contradiction sits in plain view across these stories. The sector has spent two years selling commerce signals as the antidote to cookie loss, and the appetite for those signals beyond their home marketplaces, evident in the FreeWheel integration, confirms the demand is real. The Prime Day figures are the corrective. A targeting improvement cannot fix an out-of-stock product, and the value of a precisely placed impression collapses the moment the item behind it disappears, a dependency that pure media metrics rarely capture. Underneath both stories runs the measurement problem the Incremental analysis quantified, in which the contribution of any single channel is hard to isolate from the rest of a campaign. Amazon's move to export its signals into streaming widens the surface where that question applies, because a shopper exposed on connected TV and converting on the marketplace crosses exactly the boundary that siloed measurement handles worst. The same commerce data that sharpens the targeting also complicates the accounting, since the touchpoints now span screens and platforms that seldom share a measurement framework, and the networks promising precision are still proving it across the channels where exposure and conversion separate.

The expansion of formats reinforces the point. Audio inventory entering a demand-side platform through an interface, bulk controls for off-site spend, and a hard cap on product titles are each small mechanical changes, yet together they show a marketplace continuously reshaping the surface advertisers operate on, often on timelines that collide with peak selling seasons.

Measurement and verification scramble to keep pace

Underneath the agentic ambitions sits an industry trying to prove that its impressions are real, viewable and safe. The trust layer showed cracks: a survey of the fracturing trust infrastructure noted that TAG certifications for Google and The Trade Desk lapsed, that LinkedIn invalid traffic hit 17.62 percent, and that political CTV spending doubled to 2.7 billion dollars as sports reshaped June budgets. New transparency tools arrived in response. IAS launched Quality Connect on June 18, giving publishers near real-time visibility into advertiser brand safety, fraud and suitability settings inside IAS Pulse, and the same firm brought episode-level pre-bid controls to Spotify podcast ads on The Trade Desk, covering 33 avoidance segments across 11 categories from July 2026.

Attention measurement deepened. Mobkoi expanded its partnership with xpln.ai past 1 billion measured impressions, with its Interscroller format delivering 150 percent above the display attention market benchmark. Location and outcome data moved closer to media: The Trade Desk integrated Adsquare real-world location data into Audience Unlimited, connecting digital campaigns to store visits and offline outcomes. Brand and sales lift converged in real time through two related launches: Cint merged brand lift and sales lift in one dashboard inside Lucid Measurement, and Basis integrated Cint brand lift into omnichannel campaigns for real-time awareness, ad recall and purchase intent. Independent cross-media measurement widened its coverage: AudienceProject added Pinterest to its SaaS platform, launching in the United Kingdom and Germany with further European markets to follow.

The verification cluster reflects an industry under pressure to demonstrate that its currency is sound at the moment automation is asking buyers to trust it more, not less. Lapsed certifications, double-digit invalid-traffic rates and a doubling of political CTV spending describe a trust layer straining precisely as more dollars flow through it. The new tools answer specific failures: real-time visibility into advertiser settings addresses the opacity publishers have complained about, episode-level podcast controls bring brand safety to a format that long lacked it, and merged brand-and-sales-lift dashboards collapse a measurement process that once arrived weeks after a campaign ended into something usable while the campaign is still live. Attention measurement, now spanning more than a billion impressions in a single partnership, offers a metric closer to whether an ad was actually seen than viewability ever did.

The throughline is legibility for machines. Real-time lift, episode-level controls and attention at scale are valuable to human planners, but their larger significance is that agentic systems will optimize against whatever they can read, so the metrics that survive will be the ones an agent can consume directly rather than the ones a person interprets after the fact. Each of these instruments measures a slice, and the systems being built upstream will act on whichever slices they can ingest, which puts a premium on measurement designed for machine consumption from the start.

Platforms rebuild their ad products

The large platforms spent the period restructuring core products, and Google made the most consequential change. Google is phasing out standalone Display Ads campaigns and folding the Google Display Network into Demand Gen, with a migration tool launching in June 2026, a shift that arrived as the company's display business marked a milestone, since AdSense turned 23 on June 18, having launched in 2003 to open a 100,000-advertiser network to publishers worldwide. Privacy mechanics are changing in Europe: Google will use IP addresses for ad measurement and personalization across the EEA, the United Kingdom and Switzerland from August 3, updating its TCF registration.

Measurement and conversion recovery dominated Google's product news. Google Ads is testing a feature that connects website tags to CRM and backend data sources to fill conversion gaps created by browser restrictions and ad blockers, and a server-to-server link will route Shopify purchases into GA4 from July 2026, recovering lost purchases while leaving attribution blind spots in place. Bidding is changing on a fixed timeline: Google Ads is rolling out Smart Bidding Exploration for Performance Max, a promotion mode beta, and a bidding target change on August 17. The company is also putting an event location to a community vote, asking site owners and SEO professionals to choose the city and timing for Search Central Live Deep Dive Europe 2026, with six cities in contention.

The other platforms rebuilt in parallel. X added Google Tag Manager integration, unified Conversion API developer tools and a real-time diagnostics dashboard to its rebuilt Ads Manager. Microsoft Advertising added job seniority to LinkedIn profile targeting, letting advertisers filter by 10 seniority levels across 29 markets. LinkedIn itself gave creators sharper feedback, splitting impressions into in-network and out-of-network reach in a new analytics metric. Reddit marked its 21st birthday with a transparency disclosure, as Steve Huffman revealed the platform blocks 23 million spam views dailyin its fight against bots and synthetic content. Supply-side economics shifted too: PubMatic began charging publishers 0.001 dollars CPM for excess inventory after an April 16 policy update, with its IQ team able to block domains for persistent overages. And the cookieless toolkit gained a proof point: New York Times Advertising ran its first data clean room deal in the fourth quarter of 2025, using Decentriq to match first-party data and outperform cookie benchmarks by 61 percent.

A single theme organizes the platform news: recovering signal lost to privacy changes, ad blockers and browser restrictions, then folding more of the buying process into automated systems. Google's tag-to-CRM testing, its server-side Shopify link and its bidding overhaul all point in the same direction, toward reconstructing the measurement that tracking restrictions removed and then handing more of the optimization to the platform's own models. Folding standalone display into Demand Gen continues a long consolidation of campaign types into fewer, more automated containers, a pattern that trades manual control for promised performance. The other platforms are rebuilding their own foundations, from X reconstructing its conversion and diagnostics tooling to Microsoft deepening professional targeting and Reddit publishing the scale of its bot problem as a trust signal. The clean-room result at the New York Times points to the alternative path, in which publishers and advertisers match first-party data in a neutral environment rather than relying on third-party identifiers, and a 61 percent improvement over cookie benchmarks is the kind of figure that moves the cookieless conversation from theory toward practice.

Privacy mechanics tie the platform news together as much as automation does. Routing identifiers from server to server, using IP addresses for measurement in Europe, and matching data in clean rooms are all responses to the same erosion of third-party tracking, and each shifts the work of measurement closer to first-party systems that platforms and publishers control. The direction is consistent even where the methods differ, toward reconstructing signal inside owned environments rather than borrowing it from the open web.

The business of agencies, and the wider economy

Beyond the platforms, the period exposed the financial and operational pressures shaping who can compete. A week-before-Cannes roundup captured the macro mood, with the 1.3 trillion dollar ad forecast, the completion of Vox Media's breakup, and Google's August 3 EEA deadline all landing at once. A separate stack-overhaul roundup tracked Uber taking first-party signals offsite to Meta and Google, IAS adding Spotify podcast controls, and Hyundai testing SSP bidding, evidence of advertisers pushing their own data outward rather than waiting for platforms to act.

Cash flow is squeezing the middle of the market. The freelancer payment trap described agencies facing 120-day client terms while obliged to pay freelancers within 30 days, creating a debt cascade that hits contractors hardest, per a Kaplan Group report. Tooling is attacking the administrative drag: Guideline linked its MediaTools to Prisma by Mediaocean through an expanded API, automating plan-versus-actual reporting and eliminating manual reconciliation. Workforce expectations are shifting around transparency and AI: a G-P survey of 4,000 workers across six countries found 82 percent want pay transparency while only 34 percent have it, and 32 percent of German respondents trust AI over HR for pay equity.

The infrastructure debate reached the data center. Amazon reported its data centers hit 0.12 liters per kilowatt-hour in 2025, seven times better than rivals and 75 percent toward water positivity by 2030, while the water consumed per AI token remains unmeasured. Connectivity benchmarks closed a chapter elsewhere: Australia's final NBN broadband reportfound fixed-line speeds reaching 99.4 percent of plan in March 2026, ending an eight-year consumer transparency program from the ACCC. Regional markets quantified their own digital shift: IAB Polska's 2025/2026 strategic reportshowed Poland's digital advertising market surpassing nearly 11 billion PLN, driven by programmatic, SEM, mobile and AI, while an IAB Slovenia 2026 panel reported 81 percent of companies advertising primarily or entirely through digital channels, with AI and SEO driving the biggest investment jumps. And in a commencement address that doubled as a mood reading, Sundar Pichai told Stanford graduates to sidestep AI anxiety and apply three personal filters: optimism, hard work and doing what genuinely excites them.

The financial picture underneath the technology is less buoyant than the forecast suggests. Extended client payment terms set against fixed obligations to freelancers push risk down the supply chain to the parties least able to absorb it, a structural squeeze that automation does nothing to relieve. The administrative tooling arriving to kill manual reconciliation addresses a real cost, but it also signals how much agency margin still goes to plumbing rather than strategy. The macro forecast of 1.3 trillion dollars rests on AI investment offsetting geopolitical headwinds, a balance that holds only as long as the investment keeps converting into growth rather than into stranded experiments. Regional data adds texture to the headline, with markets such as Poland and Slovenia documenting a decisive move to digital-first and AI-led spending, evidence that the shift is not confined to the largest economies. The data-center and connectivity figures, from water efficiency that still cannot be expressed per AI query to broadband speeds that finally meet their promised levels, are reminders that the physical substrate of all this software carries costs and limits the advertising layer rarely accounts for.

The mood readings around the period, from a commencement speech urging graduates past AI anxiety to surveys showing workers trusting algorithms over human departments on questions of fairness, register a workforce recalibrating its relationship to the same technology reshaping the ad business.

Out of home, and the campaigns that cut through

Out-of-home advertising kept moving toward programmatic and toward purpose. i-media integrated 1,200 motorway screens with VIOOH's SSP, offering buyers 2.9 billion monthly impressions and live ANPR-based audience intelligence across United Kingdom MSAs. Community-driven creative had its moment too: Clear Channel Outdoor partnered with Rainbow Labs to let LGBTQ+ youth in Los Angeles design Pride billboard campaigns through a structured workshop series.

Two campaign stories showed the range of what breaks through, and what crosses the line. AXE revealed that the mystery figure at the FIFA World Cup 2026 opening ceremony was a stunt for its Fine Fragrance line, a reminder that the brand, founded in 1983 and sold in more than 90 markets, still leans on spectacle. At the other end, accountability arrived for a misleading message: WeFlex paid 19,800 dollars after the ACCC found a Facebook ad falsely told NDIS participants that personal training services were automatically covered. App-economy plumbing mattered for performance marketers as well: Freecash returned to the Apple App Store on June 17, ending a two-month iOS absence that had disrupted rewarded user acquisition for mobile game publishers.

Out-of-home illustrates the same programmatic pull reshaping every other channel, with motorway inventory now transacted through a supply-side platform and enriched with vehicle-recognition data, turning physical screens into addressable media. The campaign stories bracket the period's range, from spectacle engineered to dominate the largest live audience on earth to a modest penalty that nonetheless marks a clear line on what advertising may claim. Between the giant stunt and the small fine sits the ordinary machinery of the industry, the app-store dependencies and rewarded-acquisition flows that determine whether a performance campaign can run at all, and the two-month absence of a single rewards platform was enough to disrupt user acquisition for an entire category of mobile publishers. The accountability thread runs here too, since the penalty for a misleading social ad and the transparency now demanded of out-of-home audiences both point to an industry being held to account for what it claims and whom it reaches, even as its surfaces multiply.

One pattern under the whole feed

Read end to end, the period rhymes. Platforms are shipping connectors, autonomous bidding engines and new ad surfaces now, while the audits, attribution fixes, data hygiene and legal guardrails that would make them dependable are still being drafted. The agent layer assumes clean inputs that the discovery research shows are polluted and mismeasured. The streaming buildout promises accountability while multiplying the surfaces that must be reconciled. The retail media expansion sharpens targeting while widening the attribution gap, and the regulators are arriving from several directions at once. The 1.3 trillion dollar forecast is the backdrop, and the contest at Cannes Lions 2026 is less about whether automation arrives than about who controls the interfaces it runs on, and whether the instruments to check it are built before the money fully commits.

None of these tensions is fatal, and none is entirely new. Automation has been encroaching on media buying for two decades, measurement has always lagged the channels it tries to capture, and regulators have always trailed the technologies they govern. What the period before Cannes Lions 2026 made unusually visible is the pace and the simultaneity, the sense of every layer of the stack committing to agent-driven decision-making in the same few weeks, from the connectors and the bidding engines to the streaming surfaces and the retail signals. The platforms shipping that future are betting that the apparatus to verify it will follow. The publishers losing traffic to AI summaries, the brands watching demand vanish into branded search, and the buyers handing spending to systems they cannot fully inspect are the parties who will learn first whether that bet holds. The numbers point up and to the right. The question the feed leaves open is not whether the industry will automate, but whether it can see clearly enough to trust what the automation does.