00The week in one paragraph
This was the week the machines got their assignments and the humans got their notice. OpenAI switches on ChatGPT ads across 31 European markets on 24 August — six months after its first US test, chasing $100 billion in annual revenue within four years on a surface that also decides whether to cite your reporting — and it launches contextual-only under GDPR, which quietly validates the one signal publishers own natively. Agentic buying reached audio, with Butler/Till and iHeartMedia running the first agent-to-agent streaming campaign — no SSP in the chain, CPMs 42% below the direct-buy benchmark — while the IAB Tech Lab ended the polite fiction that the two agentic standards are complementary layers: AAMP and AdCP overlap on 13 business functions, and the semantics war is now official. The referees kept consolidating — DoubleVerify’s CEO framed the $2.15 billion Nielsen deal as a single currency scoring audience and environment together — and the buy side kept industrialising, with Omnicom moving 468+ of the staffers who built its Omni platform to contractor Endava inside a $1.5 billion savings programme. Germany’s cartel office forced Apple to rewrite its ATT consent prompts after finding it held rivals to a standard it spared itself. YouTube redefined the public view (first frame, no minimum watch time, from 24 August) while keeping payment keyed to engaged metrics — and was separately reported offering creators millions to stay off Netflix. The through-line for publishers: every buyer, referee and platform is rebuilding itself around machine legibility — the inventory, consent flows and evidence that machines can read are the ones that get paid.
01ChatGPT ads land in 31 European markets — six months in, OpenAI goes for scale
Starting 24 August, ads appear inside ChatGPT across 31 European markets — Germany, France, Spain, Italy, Poland, the Nordics and more — completing something close to a global footprint six months after the first US test in February. The scale claims are not subtle: 900 million weekly active users, roughly 20% of queries carrying direct commercial intent, and a stated target of $100 billion in annual revenue within four years — a mark Meta needed 17 years to reach. The launch is deliberately constrained: ads run only on the Free and Go tiers, are withheld from under-18 accounts, and — because GDPR requires explicit consent for personalisation — arrive without behavioural targeting. Buying runs through OpenAI’s sales team and six agency networks — Publicis, Omnicom, WPP, Havas, Dentsu, MediaPlus — with self-serve to follow; in the US, the $50,000 self-serve minimum has already come down and cost-per-click now takes most of the spend (Digiday, plus OpenAI’s own announcement).
Publisher takeaway
The structural fact is that citation and paid placement now share a surface: the assistant that decides whether to cite your reporting also sells the slot next to the answer, and the balance between the two will be set by the incentives of a company on a $100 billion clock. Three moves this quarter. First, baseline your ChatGPT referrals now, before ads distort them — assistant-referred sessions are small but unusually high-intent, and once paid units compete for those clicks you will want a clean before/after rather than a guess. Second, treat citation presence as an acquisition channel with a named owner, measured monthly, the way SEO once was. Third — and this is the part worth actual optimism — Europe’s launch is a contextual-only launch: relevance from the conversation, no behavioural profile. That is the argument publishers have been making for their own first-party contextual inventory since signal loss began, about to be validated at scale with someone else’s money. Package your taxonomy, page intent and declared-audience signals so buyers hear the same pitch from you, with better provenance. And when self-serve opens in Europe, read the entry prices and CPC ranges as your earliest benchmark for what buyers think AI-conversation placement is worth against your premium slots.
02Germany makes Apple rewrite its tracking prompts — the first real crack in ATT
On 17 August, Germany’s Bundeskartellamt closed a proceeding opened in June 2022 by declaring binding a set of Apple commitments to change App Tracking Transparency. The finding, under Section 19a of the German Competition Act and Article 102 TFEU, is the one app publishers have alleged for five years: Apple held third-party apps to a consent standard it did not apply to its own services. Two changes carry the weight: Apple must align its own consent prompts with those imposed on third parties — neutral wording, symbols and layout, with room for publishers to explain the value exchange inside the flow — and publishers may combine Apple’s prompt with the consent requests data-protection law already requires, collapsing the dialog stack. Apple has four months to implement; the commitments run seven years with independent monitoring, and the binding scope is Germany only — though France has already fined Apple €150 million and Italy €98.6 million over the same framework (VideoWeek; Bundeskartellamt press release, 17 August).
Publisher takeaway
Do not expect German opt-in rates to snap back to 2020 — ATT survives, and five years of “Ask App Not to Track” trained reflexes that neutral wording only partially undoes. The real wins are narrower and worth taking deliberately. The combination right attacks consent fatigue directly: one designed consent moment, value exchange stated up front, instead of a gauntlet of dialogs — treat that flow as a product surface with an owner, not a compliance checkbox. Instrument opt-in rates now, segmented by app and cohort, because you cannot claim the uplift to your board or your demand partners without a pre-Apple-change baseline. As addressability improves, re-price German IDFA-consented supply explicitly rather than letting it blend into blind pools. And clock the precedent, because it is bigger than Apple: a regulator has formally treated consent UX — wording, symbols, layout — as an instrument of market power. Every gatekeeper consent framework that suppresses your addressable audience now has a second legal avenue pointed at it, and three European authorities converging on one framework is how a national remedy becomes a continental norm. Feed this decision into your trade-body submissions; it is the citation.
03Who counts an impression no human saw? Agent-facing ads hit their measurement wall
Digiday’s 20 August media briefing laid out the commercial half of the fight we covered last week when Perplexity blocked Time’s markdown ads: even where agent-facing formats are permitted, nobody agrees how to count them. An agent-facing ad is retrieved by a bot, folded into a generated answer, and may shape a purchase — without the person ever seeing the creative. No impression, no viewability, usually no click; agents draw on dozens of sources per answer, so attribution borders on impossible. The workarounds are inventive and mutually incompatible: OpenAds injects unique referral codes at scrape time and hopes they “survive the retrieval”; Oasy pursues CPC and cost-per-referral via dynamic insertion at the CDN layer; Time measures its markdown programme in Mobian’s proprietary visibility scores; the IAB has attribution frameworks in draft. Until something converges, advertisers are funding the category from experimental budgets only (Digiday, 20 August).
Publisher takeaway
A market can survive a format war; it cannot survive without a currency — and right now every seller in this category is grading its own homework. Publishers have lived this before, in early programmatic, and the ones who prospered backed standards and audits early rather than riding the ambiguity. But there is a structural difference this time that favours you: every attribution scheme in play runs through the publisher’s page. The referral code is injected into your content at scrape time, the CDN-layer ad rides your infrastructure, the visibility score is computed over your markdown. You are the substrate — which means data access, log rights and audit rights are negotiable now, while vendors still need your inventory to prove their models, and will be far less negotiable once a winner emerges. So contract for raw event logs, not just scores, from any agent-monetisation partner; instrument agent traffic at your own CDN and server layer, because that first-party dataset is both audit trail and future product; run parallel measurement on any live test so no result rests on the vendor’s number alone; and get into the IAB drafting process while the units your future revenue will be denominated in are still wet ink. Book whatever revenue arrives as experimental, because the buyers are.
04IAB redraws the AI disclosure line — label what changes reality, not every use of AI
The IAB published Version 2 of its AI Transparency and Disclosure Framework on 18 August, seven months after the original — and the interval is the story: since January, AI disclosure went from proposal to enforcement, with New York’s synthetic performer law live in June and both California’s SB 942 and Article 50 of the EU AI Act enforceable from 2 August. The framework’s core line is workable: disclose when AI materially affects authenticity, identity or representation — fully generated imagery and video, synthetic avatars, digital twins, chatbots a consumer could mistake for a human — and leave routine post-production, internal workflows and AI-assisted text alone. US campaigns get a standardised sparkle icon or text label. The IAB argued explicitly against over-labelling — label fatigue teaches consumers to ignore labels — and the research supports the line: clear disclosure would maintain or increase purchase likelihood for 73% of Gen Z and Millennial consumers, while 83% of advertising executives report using AI in creative, up 23 points from 2024 (IAB, 18 August; Marketing Dive).
Publisher takeaway
Read the framework twice, because publishers hold two positions in it. As content producers: AI-assisted articles, synthetic voiceovers, generated imagery in editorial and — squarely in scope — branded content studios all face the same authenticity test, and sorting every use into the framework’s disclose and no-disclose lists is an afternoon’s work that most operations will pass. Knowing that in writing is the point. As the medium of record: when a non-compliant AI ad runs on your pages, the regulator’s screenshot has your masthead on it, and compliance obligations in this industry have a long history of rolling downhill from buyers to the sell side by contract clause. Three actions: inventory your AI surface against the two lists with editorial, studio and ad ops each mapping their own; adopt the labelling standard now — one house standard aligned to the framework is cheaper than four per-jurisdiction retrofits under enforcement timelines; and add undisclosed synthetic media to your creative review criteria and your IO warranties, deciding who carries the cost of a slipped creative before anyone has an incident. The transparency high ground is also commercially real — the party with the clearest labelling wins the argument when a format is challenged, a lesson last week’s Perplexity–Time fight taught in public.
05IAB Tech Lab says the AAMP-vs-AdCP layer map is wrong — the standards actually compete
The tidy story about agentic advertising standards took a direct hit on 19 August, when Shailley Singh, IAB Tech Lab COO, published a pointed clarification: AAMP — the Agentic Advertising Management Protocols — is not an “impression-layer protocol” sitting neatly beneath AdCP’s campaign planning. It is an umbrella framework spanning planning, discovery, transaction management and impression-time execution — and by the Tech Lab’s own accounting, AAMP and AdCP overlap on 13 business functions, from natural-language briefs to deal creation. Singh also dismantled the softest complementarity argument: both stacks speaking MCP proves nothing, because “MCP is a communication substrate; it does not reconcile their advertising semantics.” AdCP launched last October with six founding members — notably without Google, Amazon, The Trade Desk or Microsoft — and some players, like DOOH seller VIOOH, have hedged by planning to register under both (IAB Tech Lab, 19 August; PPC Land).
Publisher takeaway
We covered AAMP 2.3 in our 2 August issue as agentic buying getting real guardrails; this week the standards body itself ended the polite fiction of complementary layers. That changes the advice. “Implement both, they’re different layers” was cheap counsel when the layers were believed distinct — thirteen overlapping functions means dual implementation is duplicate integration work against semantics that may diverge further. The question to ask of every hour of agentic engineering spend: does this survive either outcome? The work that does is the work that is about your house rather than either protocol’s semantics — clean, machine-readable inventory declarations, accurate product catalogues, honest pricing structures, reliable reporting. Every buyer agent needs those under every standard; front-load them. Pilot narrowly where a counterparty you actually trade with is transacting today, and keep contract language free of commitments to either standard’s representation of your inventory while the overlap is unresolved. Then make your vendors show you the map: which functions they implement under which protocol, and how they will handle divergence. “We support both” without specifics means the reconciliation cost is being deferred — most likely to you. The war is over the semantic layer, not over whether agentic buying happens; prepare for the second and let the standards bodies fund the first.
06The quiet yield leaks in programmatic CTV — publishers leaving money on the table
Kieran Greene, founder of sell-side firm Shinka and formerly of Google, told VideoWeek on 19 August that the plumbing between CTV publishers and bidders routinely underprices inventory — and that publishers typically see a 20–40% revenue uplift once bid-path inefficiencies are fixed. A vendor’s number, but the leaks he catalogues are checkable: bid requests going out malformed or missing the datapoints that decide whether an advertiser bids at all — show, season, episode, ad-break position; geographic signals arriving half-built (a ZIP code with no city or country attached); translation quality varying by demand partner, so where the work is skipped, demand simply fails to appear. Fewer bidders seeing a well-described impression means lower bid density and CPMs settling below what the inventory would clear in a fair contest. In DOOH, trading through the same pipes, audience multipliers get lost entirely — “an ad seen by fifty people in a bar or forecourt is often traded as if it were seen by one person on their sofa” (VideoWeek, 19 August).
Publisher takeaway
What makes this class of problem expensive is that it is invisible by design: a malformed request does not error out, it just attracts fewer and lower bids, and no report logs the bids that were never made. Your yield dashboard shows a healthy auction while underpricing every impression in it. The response costs days, not a replatform. First, audit your own bid requests before believing anyone’s uplift number — capture a sample per demand path and score them for content metadata, complete geography and pod position, correctly formatted, consistent across partners; that converts Greene’s thesis from claim to measurement on your inventory. Second, make signal completeness a commercial term: which fields each SSP passes to which DSPs, where requests get truncated, what gets enriched — and ask vertically integrated stacks specifically how third-party demand paths are serviced relative to their own, in writing. Third, measure bid density per path and per signal-completeness cohort, raise floors where enriched requests demonstrably clear higher, and reopen terms on paths that persistently deliver thin competition. Content metadata is pricing power: show-level signals are what let advertisers pay premium prices for premium context, and withholding them — deliberately or through neglect — converts your best inventory to run-of-network pricing. In a market fought over single CPM points, revenue recovered from your own plumbing is the cheapest money in the building.
07Smart TV home screens get standard specs — the premium slot outside the stream opens up
Nexxen announced standardised inventory specifications for smart TV home-screen advertising on 20 August, unifying the hero banners and tiles that greet every viewer at power-on across OEMs whose sizes, formats and click actions previously all differed. DSPs can now bid programmatically across manufacturers — the launch roster includes The Trade Desk, DV360, StackAdapt, Basis, Brainlabs and H/L, activating against V (formerly VIDAA), TCL and TiVo Ads inventory — with Nexxen Studio handling creative adaptation and onboarding time for new OEM partners cut by as much as 50%, per Nexxen. The demand logic rests on one Nielsen figure: viewers spend an average of 10.5 minutes deciding what to watch after turning on the TV — a long, attentive window in which the home screen is the only ad surface in the room. The buyer mix is already shifting from entertainment tune-in budgets toward performance advertisers (ExchangeWire, 21 August; AdExchanger, 20 August).
Publisher takeaway
The strategic fact is uncomfortable and better confronted early: the home screen sits above every app in the device hierarchy, and once it is standardised and pitched to performance budgets, it competes for the same advertiser dollar as your in-stream inventory — with a reach story (every session, whichever app wins the viewer) that app-level sellers cannot match on that surface. It is not zero-sum: home screens are still largely a discovery format, and a liquid tune-in market drives audiences into apps where you monetise the actual viewing. The nearer risk is subtler — ease of transaction has redirected spend before, independent of media quality, and a one-upload, standard-spec buy is now the frictionless option. So respond with symmetry. Audit how your CTV inventory reaches DSPs — deal setup time, creative specs, activation friction — against that bar, and fix your side before budgets choose for you. If you operate a FAST channel or app, price out hero and tile placements for your own tune-in marketing while the buyer pool is thin; the publishers who understand this inventory as buyers will negotiate it best as competitors. Sharpen the case only in-stream can make — engaged viewing, content adjacency, completion — because a performance buyer comparing surfaces will otherwise compare on CPM and ease alone. And push, through IAB Tech Lab and your SSPs, for native CTV specs to live with a neutral body: one vendor’s spec becoming the de facto standard for a premium format is a governance question, and the cheapest moment to raise it is now.
08YouTube redefines the view and doubles the monetisation bar
From 24 August, YouTube counts a public view the moment playback starts, with no minimum watch time — across long-form, Shorts and live — deliberately aligning with how TikTok and Instagram count. The old standard survives as “engaged views” inside Analytics, and everything commercial still runs on it: creator earnings and Partner Program eligibility remain keyed to engaged metrics, not the inflated public counter. While the public number loosened, the earning bar rose: from 1 February 2027, new Partner Program applicants need 8,000 watch hours (double 4,000) or 20 million Shorts views in 90 days (double 10 million). Digiday’s ledger of winners and losers: established long-form channels barely disrupted, cross-platform publishers gain comparable counts, channels entering Q4 sponsorship talks get bigger toplines — while slower, higher-craft short-form creators face a volume treadmill, and measurement founder Nick Cicero warns that “YouTube is inflating the top-line number while pushing meaningful signals deeper” (Digiday, 20 August).
Publisher takeaway
State precisely what happened: YouTube forked its core metric — a social number for the scroll and the screenshot, a commercial number for the money — and payment follows the engaged fork. The risk to publishers is entirely operational, and it has a date. Every YouTube benchmark breaks on 24 August: view counts, view-through rates and every engagement ratio calculated against views become discontinuous, not because audiences changed but because the denominator grew. So restate every view-denominated agreement before the numbers fork — sponsorship guarantees, CPV buys, internal targets — specifying public views or engaged views in each, and agree the conversion basis on any deal spanning the date now rather than in a reconciliation dispute. Rebuild dashboards on engaged metrics as the primary series, keep public views as a reach line, and annotate 24 August on every chart so nobody reads the seam as performance. Then bank the two genuine gifts: with YouTube, TikTok and Instagram finally counting alike, you can build the first honest cross-platform comparison your sales team has ever had — pair the bigger toplines with engaged-metric evidence, because sophisticated buyers will discount unaccompanied inflation — and note that the doubled 2027 thresholds favour incumbents: existing monetised channels are not re-tested, and payment keyed to engagement rewards exactly what professional publishers produce. The substance is favourable; only the bookkeeping is dangerous.
09Agentic buying reaches audio — Butler/Till and iHeartMedia go agent-to-agent, no SSP in the chain
Independent agency Butler/Till and iHeartMedia completed what they describe as the industry’s first agentic streaming-audio campaign: a four-week July–August pilot across streaming audio and podcasts, on a budget just under $10,000, for an agricultural client. The architecture is the news: two AI agents — one the agency’s, one the publisher’s — negotiating directly over an MCP server, where Butler/Till’s earlier pilots ran through intermediaries like PubMatic. Reported results: streaming CPMs 42% below the client’s traditional direct-buy benchmark, and 48% of podcast impressions in non-skippable mid-roll against 33% under traditional planning. Butler/Till has now run agentic pilots with a dozen clients across CTV and display; WPP Media and Omnicom are testing comparable tooling. iHeart’s chief business officer Lisa Coffey stated the seller’s motive plainly — make audio as easy to activate as the large digital platforms — with broadcast radio opening to agentic transactions later this year and Audiograph launching through Amazon’s DSP in mid-September (Digiday, 20 August).
Publisher takeaway
Two separate lessons, both transferable far beyond audio. The channel lesson: audio has been under-bought relative to consumer time for a decade because it is hard to buy, and agentic buying just demonstrated it can dissolve exactly that friction — meaning agents are a demand-unlocking play for every hard-to-buy channel, which is the opposite of a threat for sellers equipped to transact. The path lesson: an agency system and a publisher system completed a deal with no exchange in the middle. One pilot does not retire the SSP — aggregation, decisioning and clearing do not replicate point-to-point at scale — but the demonstration stands, and it reshapes the make-or-buy question for publisher ad stacks. Note also what the machine did with the money: it did not just buy cheaper, it allocated nearly half of podcast delivery into the most premium placement class. Machine allocation follows value density where value is legible — sellers whose best placements live in a rate-card PDF and a seller’s memory should expect agents to miss them. So structure catalogues, placements, pricing tiers and availability as clean queryable feeds; make deal creation executable end-to-end by a counterparty’s system with your controls enforced in the pipe, not the inbox; decide deliberately who represents your inventory to buyer agents and at what take, per the SPO audit we urged on 16 August; and pre-model what agent-optimised buying does to your yield mix — the 42% headline will be quoted at you in negotiations regardless of its sample size, so have your own account ready before buyers supply one.
10One currency to score them all: the Nielsen–DoubleVerify deal finds its meaning
Two weeks after Nielsen’s $2.15 billion DoubleVerify acquisition, the sharpest description came from DV CEO Mark Zagorski himself: the combination creates a single currency scoring media on both audience delivery and environment quality — while Paul Evans of V2RSION skewered the companies’ own “independent media intelligence platform” framing as so generic it “could basically cover half the businesses in the Lumascape.” The combined entity carries roughly $4 billion in annual revenue and unites Nielsen’s cross-screen measurement, DV’s verification and Rockerbox’s attribution: three verdicts that used to come from separate parties, issued under one roof. The same week, Nielsen’s Brian Fuhrer detailed pre-fall-season methodology work aimed at “removing bias”: American Community Survey language data for Spanish-language measurement, deeper integration of passive wearable meters for co-viewing (piloted at the 2026 Super Bowl), and a fix for the ARF universe datasets that ran 12–15 months stale — a staleness publishers had complained about (ExchangeWire, 20 August; AdExchanger, 19 August).
Publisher takeaway
Both halves are true and deserve different responses. The methodology work is real and will move your numbers before the fall season prices: if you carry Spanish-language, bilingual or heavily co-viewed programming, ask Nielsen for the expected directional impact on your reported audiences and take it into guarantee negotiations rather than discovering it in the first fall report. And log the meta-lesson of the ARF fix: the staleness timeline moved because publishers pushed — collective pressure, applied early, changes methodology. The consolidation half needs contractual caution. A fused audience-plus-quality score is genuinely useful to buyers — one number instead of a reconciliation exercise — but for publishers it concentrates pricing power: today measurement and verification are separate line items, each negotiable, each replaceable in principle; if buyers transact on a single currency, you carry it at the integrated vendor’s price. And the deal’s economics point one direction — $2.15 billion wants earning back, and cross-selling bundles is the obvious mechanism. So get the bundling question answered in writing before the Q1 2027 close: will measurement and verification remain separately contractable at arm’s-length prices, and will the single score be available in components? A vague answer is information; a written one is leverage. Meanwhile keep the parallel evidence chain funded — log-level ad server data, independent viewability, your own IVT reporting. When one company issues three verdicts, the records you keep yourself are the only second opinion left standing.
11YouTube pays to keep its stars off Netflix — and content just learned its price
YouTube is offering major creators millions of dollars to keep their videos exclusive for set windows, Bloomberg reported on 19 August — a defensive response to Netflix’s accelerating raids: Mark Rober now central to Netflix originals, Ms Rachel’s library on the service, Alan Chikin Chow and Nick DiGiovanni signed, and Netflix also acquiring episodes of premium publishers’ YouTube shows. The counter-offer has a carrot — exclusivity payments, direct programme financing, a share of major brand deals — and a stick: YouTube has told creators that multi-homing on Netflix hurts their YouTube viewership and may cost them marketing support, event placement and brand-campaign access. No deal was finalised at the time of reporting. In the same VideoWeek round-up: India abolished its 12-minute-per-hour TV ad cap, in force since 2006, via the Cable Television Networks (Amendment) Rules — the ministry noting the market has gone from 62 channels to more than 900, while digital faces no load restriction at all (Bloomberg, 19 August; VideoWeek, 21 August).
Publisher takeaway
Strip the creator-economy framing and this is two distributors bidding against each other for the supply that differentiates them — the market condition under which licensing terms improve, and publishers with video libraries and platform-native shows are inside the repricing, not spectators. Three leverage lessons transfer. The audience relationship is the asset, and portability is what makes it priceable: Rober commands these terms because his audience follows him; a publisher with newsletters, apps and a name people seek out holds the same leverage at smaller scale, while one whose reach arrives only by algorithm holds none. Exclusivity now has a visible market price — which means non-exclusivity has one too: in every licensing negotiation, video platforms and AI retrieval deals alike, price exclusivity as a separate term rather than a default you give away inside a standard distribution agreement. And read YouTube’s stick as the template it is: reduced featuring and withheld brand access for multi-homing creators is single-platform dependency risk made explicit — the same enforcement pattern as Perplexity’s trust-score threats two weeks ago. Multi-home deliberately while terms are friendly. On India: watch the natural experiment. An entire 900-channel market just flipped the supply constraint off on a single date; the next two quarters will show whether broadcasters protect price with discipline or race to saturation. Pre-decide your own ad-load ceiling as policy — chosen for lifetime value — before competitive pressure decides it for you, and expect “digital has no cap” to surface as a deregulation argument in other markets.
12Omnicom hands the builders of Omni to a contractor — read it as a buyer-side signal
Omnicom spent the year selling its AI-powered Omni platform as the keystone of the merged Omnicom–IPG offer, showcasing it at CES in January. In June and July, per a 20 August Adweek exclusive, it quietly transferred at least 468 of the staffers who built it — mostly product and engineering — to technology services firm Endava across the US, UK, India and Malaysia, with roughly 50 US platform staff receiving layoff notices on 9 June. Omnicom frames it as a capacity-increasing multiyear partnership, with platform IP, data science and client relationships staying in-house. The context makes the economics legible: since closing the IPG deal, Omnicom targets about $1.5 billion in merger savings over 30 months — double the original ambition — with around 4,000 job cuts announced at close and outsourcing named among the levers (Adweek, 20 August; prior merger coverage).
Publisher takeaway
Resist the schadenfreude and read the chain that matters. Holdcos are consolidating spend decisions into automated platforms — Omni sits on Acxiom identity data and increasingly mediates targeting, planning and measurement for a vast share of brand budgets — and those platforms are simultaneously being engineered for cost. A buying system maintained at contractor economics inside a $1.5 billion savings programme is a system tuned for efficiency: it favours automatable supply paths, standardised measurement and algorithmic allocation, and the human planner who understood why your context was worth a manual line on the plan is precisely the cost such programmes remove. The response has three parts. Make your inventory legible to the machine that replaced the champion — clean sellers.json and ads.txt, rich contextual signals, curated packages and deal IDs an algorithmic planner can find and score — because the platform, not the planner, is now your buyer of first instance at every holdco. Rebalance sell-side effort toward direct brand relationships, which do not route through an agency cost programme at all, as a hedge against two years of structurally unstable agency headcount. And add a build-and-maintain question to your platform diligence everywhere: for each buying platform, SSP and measurement partner material to your revenue, establish who actually engineers it and under what arrangement — because if the largest holding company can hand its flagship platform’s engineering to a contractor mid-AI-race, so can the vendors you depend on. The agencies are deciding what they will stop owning; be equally deliberate about what you refuse to outsource — the audience relationship, the data, and the evidence of your own value.