Why Lapis Is Building the AdSense for the AI Era

The Thesis in One Sentence

The next great advertising company will represent demand itself. It will hold the advertiser’s objective, decide where the next dollar goes, and carry learning across every seller. We are building Lapis to be that company.

Without a buyer of its own, a business enters every platform as a beginner. Its brand memory lives in folders. Its campaign memory lives in disconnected dashboards. Its economics are interpreted by the same sellers asking for its budget. Every new channel creates another specialist, another handoff, and another reset. The largest cost is the experiment the business never gets to run.

With Lapis, the business keeps the memory. A campaign on one channel can improve the next campaign on another. A customer objection can change the next creative brief. A failed offer becomes a remembered fact instead of a forgotten report. The compounding asset belongs to the advertiser.

A business should be able to state an outcome: “Acquire 500 qualified customers at a profitable payback, without breaking these brand and compliance rules.” A system should do the work between that objective and the result. It should research the market, form a strategy, create hundreds of truthful and on-brand messages, adapt them to each surface, launch tests, watch spend, measure outcomes, stop failures, scale winners, and remember what it learned. A human should set the truth, taste, risk tolerance, and destination. Software should run the loop.

The category is an independent ad buyer: a persistent system that accepts a business outcome and delegated authority to create, allocate, act, measure, and learn across sellers. Lapis begins with the hardest missing input, creative that is specific, plentiful, on-brand, and ready for each channel. The product layers extend through planning, forecasting, and competitive intelligence. Managed engagements extend into campaign operation, measurement, and iteration. The destination is simple: one goal in, a continuously improving portfolio of advertising out.

This Is Bigger Than a New Ad Channel

Calling AI advertising the next stop after search, social, and retail media makes the future too small. AI creates new inventory inside ChatGPT, AI Mode, Copilot, and conversational commerce. It also changes the production function of advertising everywhere else. It lowers the cost of research, strategy, production, analysis, and coordination at the same time. It changes who can be a competent advertiser.

The Industrial Revolution mechanized physical transformation. The internet made information transmission nearly free. AI is beginning to reduce the marginal cost of competent cognitive work: drafting, comparing, designing, translating, analyzing, deciding, and coordinating tools. Expertise becomes more leveraged while institutions transform. The International Labour Organization’s 2025 assessment says transformation is more likely than wholesale replacement. Capabilities once bundled inside institutions can now become products available to individuals.

The early evidence points in exactly that direction. The Stanford AI Index reports that generative AI reached roughly 53% population adoption in three years, faster than the personal computer or internet, while 88% of surveyed organizations used AI in 2025. In a real-world study of 5,172 support agents, AI assistance raised productivity by about 14% on average, with much larger gains for less-experienced workers, according to the National Bureau of Economic Research. AI’s democratizing effect comes from moving useful know-how down the experience curve.

Advertising is unusually exposed because so much of the work is digital, repetitive, measurable, and language-or-image based. The global allocation problem is also enormous: WARC forecasts $1.30 trillion in worldwide ad spend in 2026. The opportunity is to make world-class decisions about that trillion-dollar pool available to every business.

The AdSense Moment Has Arrived Again

The name is deliberate. AdWords opened digital advertising to the long tail of demand. AdSense opened it to the long tail of supply. Together, they turned a market that required specialist relationships and technical infrastructure into a market that any advertiser or publisher could enter through Google.

Google’s own 2008 founders’ letter says its first self-service AdWords system launched in 2000 with 350 advertisers, helped democratize access to advertising, and paired with the publisher-facing AdSense program. Google’s official AdSense history dates its contextually targeted text ads to 2003. In operation, a publisher pastes code on a site, advertisers bid in a real-time auction, and Google bills the advertiser and pays the publisher. The network now pays billions of dollars a year to publishers.

AdWords opened demand. AdSense opened supply. Lapis opens intelligence. The media platforms will own their audiences, inventory, and auctions. Lapis will own the portable understanding and operating loop that lets an advertiser win across them. The same four forces that made Google’s participation layer historic now apply to the advertiser’s decision layer:

• Remove the participation bottleneck. Replace relationships, specialist labor, and minimum efficient scale with a simple interface.
• Make context computable. Match a commercial message to the meaning of a page, query, conversation, or task.
• Open the market to the long tail. Let a tiny business use capabilities previously reserved for the largest buyers and sellers.
• Become the default on-ramp. Own the layer through which millions of participants enter the new economy.

AdSense standardized access to long-tail supply. AdWords standardized access to long-tail demand. Lapis standardizes and automates the intelligence required to turn a business goal into effective advertising. This is the next participation unlock, and it sits on the advertiser side.

The First Ad-Tech Revolution: Brilliant Companies, Incomplete Control Points

The history of digital advertising is often told as a sequence of acronyms: ad server, network, exchange, DSP, SSP, DMP. The more useful way to read it is as a sequence of bottlenecks. Each generation produced an excellent company that removed one bottleneck. Most were then absorbed by a company that controlled something scarcer.

These companies were important enough to buy. Their limitation was structural: even the broadest independent stacks controlled too few of the scarce assets that turn technical capability into durable distribution advantage. Those assets included advertiser demand, publisher inventory, consumer attention, first-party data, an operating system, and the client budget.

DoubleClick is the clearest example. The FTC’s 2007 description noted that DoubleClick delivered and reported ads after advertisers and publishers had agreed to terms; it did not itself buy or sell the media. Google supplied the missing demand. A 2025 federal court opinion found that DoubleClick’s publisher server had about 60% share when acquired and that Google’s DFP later handled 91% of worldwide open-web publisher-ad-server impressions in 2022. The court held that plaintiffs failed to show the DoubleClick or Admeld acquisitions were anticompetitive when viewed in isolation. It separately found that the later tie between DFP and AdX was unlawful under Sections 1 and 2 of the Sherman Act and that related conduct unlawfully maintained monopoly power. The lesson is direct: infrastructure becomes extremely durable when joined to unique demand and distribution.

Why the Big Six Survived Programmatic

The “Big Six” holding companies were Publicis, WPP, Omnicom, Havas, IPG, and Dentsu. Their media agencies bought enormous volumes on behalf of brands; their creative and specialist agencies supplied the rest of the campaign. The label is now historical. Omnicom completed its acquisition of IPG in November 2025, turning six into five. The operating model matters more than the count.

Programmatic was supposed to remove manual buying. It did, at the impression level. It did not remove the buyer. Agencies retained five things the tools did not supply:

1. The client and budget relationship. The brand hired the agency to represent it; the agency selected and operated the tools.
2. The translation layer. Someone still had to turn a business objective into a media plan, an audience, an offer, a brief, a set of assets, and a measurement plan.
3. Creative production. Auctions could select among ads, but they could not invent a genuinely new, brand-safe campaign every time the market changed.
4. Organizational accountability. Global brands needed approvals, policy review, procurement, reporting, and a person to call when something failed.
5. Complexity management. Every new DSP, SSP, identity vendor, verification service, private marketplace, and measurement product created another specialty for the holding company to staff.

The incumbents also internalized the disruptors. WPP bought 24/7 Real Media, built Xaxis on its technology, and invested in AppNexus. Agencies created trading desks and data teams. The technology became part of their bundle. In one sentence: ad tech did not kill the agency; it gave the agency a trading desk.

The result was automation without simplicity. In a sample of 21 advertisers’ open-web programmatic activity covering $123 million in spend and 35.5 billion impressions, the ANA’s 2023 supply-chain study classified 29% of each dollar entering a DSP as transaction costs. It classified another 35% as “loss of media productivity” from non-viewable and invalid traffic plus non-measurable and made-for-advertising inventory. The report says each advertiser must decide the value of non-measurable and made-for-advertising inventory. Under the ANA’s TrueAdSpend definition, 36 cents of each dollar effectively reached the consumer. Agency fees and brand-safety costs sat outside the study. The machinery got faster while the system got harder to understand.

Brands were already bringing more marketing capability in-house. In a 2023 survey of 162 ANA members, 82% reported an in-house agency, up from 42% in 2008. Among the in-house agencies, 54% handled some media planning or buying, while 92% of respondents still used external agencies. Control was already moving closer to the advertiser. AI accelerates that direction because it turns capacity and specialist execution into software.

AdWords and AdSense made large institutions optional for participation. Lapis makes agency-grade execution directly accessible, then turns it into the default layer through which the long tail buys. Google proves the durability of demand ownership. The Big Six prove the durability of the brief and budget relationship. Lapis combines those positions in software.

The Break: From Automation to Agency

Old ad technology could choose from options humans had already made. Generative and agentic systems can create new options, evaluate them, take actions through tools, observe the result, and choose the next action. That is the boundary between automating a step and automating an operator.

This is why the present shift is different from DoubleClick, mobile, or real-time bidding. Those changes moved a known object faster. AI can construct the object and operate the process around it. At far lower cost and with far more generality, strategy, creative, campaign setup, analysis, and the next decision can now enter the same feedback loop.

Seven Laws of AI-Era Advertising

Calling the new unit a “conversational display ad” understates the change. A small sponsored card is only the first implementation. The deeper primitive is a governed commercial object that can answer, adapt, offer, and eventually transact. Seven changes follow.

1. The unit of intent expands from a keyword to a situation

A search query might say “best CRM.” A conversation can reveal a 12-person agency, a Salesforce migration, a $500 monthly ceiling, a requirement for EU hosting, frustration with implementation time, and a decision deadline next week. It is a live description of the job the buyer is trying to complete.

2. Meaning matters more than exact match

OpenAI’s current advertiser documentation says ChatGPT primarily selects ads by relevance to conversational context and intent, using context hints, the landing page, title, and copy in a relevance-weighted second-price auction. Keywords do not disappear, and identity signals do not vanish everywhere. But the center of gravity moves from “who has this cookie?” or “who typed this string?” toward “what is this person trying to accomplish right now?”

3. Inventory compresses

A search page can show many links and several ads. A feed can create effectively infinite slots. A trusted answer has room for very few. OpenAI currently shows a separate sponsored unit below eligible responses; Google’s AI Mode tests use a small number of highlighted or conversational placements. Fewer slots make each decision moment scarcer and potentially more valuable while making each irrelevant message more costly. Relevance becomes the price of admission.

4. Creative becomes generative and adaptive

Web-era dynamic creative inserted a product image, price, or city into a template. AI-era creative can change the explanation itself. Google is testing AI Mode formats in which Gemini builds query-specific creative and an independent explanation of why an offer may fit. It has also announced AI-powered Shopping ads, Business Agent for Leads, promotion bundling, and checkout using UCP. The campaign becomes a body of verified facts, brand rules, assets, offers, and permissions from which a platform can generate the right expression.

5. The destination collapses into the interface

The old ad earned a click to a page where the real work began. AI interfaces are pulling comparison, qualification, conversation, and checkout into the same surface. OpenAI and Stripe’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol define standardized ways, and are beginning to power live experiences, in which agents and merchants exchange product, order, and payment instructions. As the distance from discovery to action approaches zero, completed value becomes the governing metric.

6. Machines appear on both sides of the market

The consumer may delegate research, comparison, monitoring, or purchase to an agent. The business will delegate campaign creation, testing, and buying to another agent. The ad of the future must therefore persuade a person and survive machine scrutiny. Claims need evidence. Product data must be structured and current. Price, inventory, eligibility, and terms need to be executable. The winning message cannot merely be loud; it must be legible and useful to another system.

7. Trust moves from the policy page into the product

AI conversations can be private, sensitive, and advisory. That makes covert influence intolerable. OpenAI separates ads from answers, withholds conversations from advertisers, gives users controls, and excludes sensitive contexts. Anthropic has taken the other defensible position and declared that Claude will remain ad-free. The market will contain both models. Either way, commercial systems win only when provenance, labeling, consent, claims, and user control are built into the unit.

The Market Has Already Turned

The future is already shipping, although its pieces sit at different stages and use different business models.

OpenAI’s scale makes the opening material. In February 2026, OpenAI reported more than 900 million weekly active ChatGPT users. That figure is total platform reach. The advertising opportunity began with eligible Free and Go users in the United States, while paid plans remained ad-free, and expansion has been phased. In May, OpenAI announced agency and technology partners plus a gradually opening beta Ads Manager. Partners can support budgeting, bidding, and creative while OpenAI controls delivery. The named agency partners included Dentsu, Omnicom, Publicis, and WPP. The pattern is plain: the new surface exists, the incumbents entered first, and an external partner ecosystem is expanding.

Fragmentation multiplies the value of a neutral buyer. Claude can remain ad-free. Other experiments can retreat. Traditional Google, Meta, Reddit, LinkedIn, retail media, streaming, and open-web inventory will continue. Every channel is becoming more automated while commercial discovery moves into AI-mediated moments. Lapis follows the advertiser’s outcome across whatever surfaces earn attention.

Creative Is the Wedge. The Buyer Is the Company.

Creative is the first binding constraint and the opening wedge.

An optimizer is only as useful as the options it can test. A team producing four generic assets per quarter gives bidding intelligence almost nothing to discover. Conversational advertising makes the action space much larger: distinct needs, objections, products, audiences, languages, formats, placements, offers, and stages of decision each deserve a truthful variation. OpenAI’s own creative guidance tells advertisers to build for coverage with a high volume of diverse ads and genuinely different angles. The old workflow, where every variation requires a new brief, copy round, design round, resize, review, and upload, collapses under that requirement.

Generative creative removes that first constraint. More importantly, it creates the action space an agent needs in order to learn. A system can test a value proposition against a use case, observe the outcome, generate a better version, and carry that learning into the next channel. Creative stops being a warehouse of assets and becomes a controlled policy for expressing the brand under different conditions.

That is why “one prompt makes an ad” is a useful demo but an incomplete company. The durable product must know what is true about the product, which claims are allowed, how the brand should feel, what each audience values, what has already been tried, which outcomes were incremental, and what the system may do next. Creative opens the loop. Memory, action, measurement, and governance close it.

The wedge is a permission ladder

Creative delivers immediate, inspectable value before the customer hands an agent meaningful budget authority. First Lapis earns permission to learn the brand and propose work. Then it earns permission to publish or export, observe outcomes, make bounded changes, and allocate spend. Each step produces the proof required for the next. Lapis earns trust before it receives control.

AdSense began with a snippet, not a demand that publishers replace their entire business. Lapis begins with an output a marketer can inspect and approve. More truthful variants create more experiments. More experiments create outcome memory. Outcome memory improves allocation. Better allocation earns more budget. More budget deepens the memory. That loop is how a creative wedge becomes the buyer.

What Lapis Is Building

Lapis is building the advertiser-side operating system for this loop. The product today already exposes the architecture in layers. Each layer expands toward the full autonomous buyer.

The product ladder maps the business strategy. Self-serve plans land through an immediate workflow: understand the brand, plan the campaign, create variants, forecast, and prepare the work. Managed plans expand that relationship into launch, operation, reporting, and iteration across supported channels. Every step increases the frequency of use, the depth of context, the responsibility Lapis can assume, and the value it creates. A creative tool can be used occasionally. A buyer operates every day.

The wedge is already in the market. Lapis’s Y Combinator profile reports usage by more than 1,000 marketing teams. The live G2 profile shows a 5.0 rating across 135 reviews. The reviews give the abstraction concrete shape. One enterprise leader reports producing more than 200 localized LinkedIn assets in under two hours. An agency founder reports turning a brief into 20 platform-ready ads inside 45 minutes. A healthcare executive reports moving from a product update to live Meta or LinkedIn creative in under 10 minutes. These user-reported workflows show the same thing: Lapis collapses production time while preserving brand context.

I came to this problem after building sales and marketing functions at Warp and helping take the business from zero to $2.5 million. My co-founder Sai Surbehera previously built recall and reranking systems for Walmart Search. One side of the founding team understands the advertiser’s operating problem. The other understands how intelligent systems rank and select among competing options. We founded Lapis at that intersection.

Why Lapis Has the Right Sequence

Lapis enters before the auction, through a frequent and visible pain: turning a real business into specific, on-brand advertising. That position lets Lapis learn the product, audience, voice, visual language, claims, offers, and constraints that every downstream platform needs but no media seller can own neutrally.

The sequence is deliberate. Brand context comes first. Creative turns that context into testable hypotheses. Managed execution produces outcomes. Outcomes create experiment memory. Experiment memory earns the right to recommend allocation. Each layer makes the next layer safer, more useful, and harder to replace. By the time Lapis moves a dollar, it can already know why the campaign exists, what the business may say, what has been tried, and what result actually matters.

A platform optimizer begins inside one auction. A DSP begins near the bid. An agency begins by assembling people. Lapis begins with the business itself, then follows the objective into every execution system. The strategic position is upstream of all three. Lapis captures advertiser state when it produces an approved commercial output, then moves downstream into measurement and allocation.

Every rival must cross a structural boundary to reach that position. A platform must become willing to route spend away from itself. A DSP must move upstream to before the campaign exists. An agency must convert labor and client memory into product. A marketing cloud must become a system of delegated action. A generic model must earn permissions and build vertical integrations. Lapis begins at the intersection where those paths converge.

Why the Neutral Buyer Wins

Google, Meta, Amazon, Microsoft, and OpenAI are all building increasingly capable creation and optimization tools. Any thesis that assumes the platforms will stop at media delivery is already obsolete. Google Performance Max uses AI across bidding, budgets, audiences, creative, and attribution. Meta Advantage+ automates audiences, placements, budgets, and creative. Amazon has Creative Agent and Ads Agent. OpenAI is expanding its own advertiser tooling.

Platform agents will dominate local optimization inside their own auctions. The control point Lapis is building sits above them: which platform deserves the advertiser’s next dollar? Seller-owned systems become execution engines while Lapis preserves the objective that connects them.

A seller-owned optimizer sees its inventory and its attribution. It is rewarded for producing value inside that boundary and for keeping spend there. A buyer-owned system compares opportunity cost across sellers, keeps the brand consistent across them, reconciles results with permissioned first-party business outcomes, and moves spend from one platform to another. This is the ordinary difference between a media seller and an advertiser’s representative.

Neutrality has a concrete product definition: portable advertiser-owned context, channel-agnostic success metrics, transparent economics, auditable allocation decisions, and the ability to recommend less spend. Independence from a media owner creates the possibility. The buyer earns the label through its incentives and behavior.

Neutrality also makes the system durable when surfaces diverge. ChatGPT may run ads. Claude may not. Google may synthesize the unit. Meta may render dozens of variants. Amazon may complete the transaction. The brand still needs one portable layer that knows what is true, what it is allowed to say, what success means, what has worked elsewhere, and how much autonomy to grant. Platforms render and deliver. Lapis is built to own the advertiser’s continuity.

Why this is different from old neutral ad tech

Neutrality alone is not a moat. Position is. Admeld and AppNexus sat in the middle of the market and rented access to demand and infrastructure. Google could withhold unique demand, privilege its exchange, and control the publisher server around them. Lapis begins upstream, where advertiser demand originates. Platforms can restrict inventory and APIs. They cannot withhold the advertiser’s own objective, product truth, brand rules, permissioned outcomes, or accumulated experiment history.

If one platform narrows an interface, Lapis loses an execution route, not the customer’s objective or accumulated state. Customer-owned accounts, channel-ready creative, managed workflows, first-party measurement, and the next decision remain valuable. The auction can change without erasing the advertiser record. That is the difference between a neutral intermediary and an advertiser-owned control point.

The Moat Is the Loop

A foundation model can reproduce an image. It cannot reproduce a business history it never observed. Lapis’s moat is the integration of five assets that become more valuable together and with use.

1. A brand graph. Products, audiences, voice, visual language, approved claims, prohibited claims, references, prices, offers, legal rules, and past decisions remain available across campaigns.
2. An experiment memory. The system remembers winners, the hypothesis behind each test, the conditions, the audience, and the confidence in the result.
3. A cross-channel outcome graph. It connects creative and spend decisions to permissioned customer outcomes such as leads, purchases, margin, or payback rather than accepting each platform’s isolated view as the whole truth.
4. An action surface. Integrations, schemas, launch workflows, approvals, pacing, monitoring, and reporting turn intelligence into reliable work. Models suggest; products act.
5. A trust system. Permissions, budget caps, policy checks, audit trails, human review, and stop conditions make autonomy safe enough to use with a real brand and real money.

The five assets become one flywheel

• The advertiser loop. More use produces richer brand context, more experiment history, clearer constraints, and better outcome calibration. Leaving means rebuilding decisions, approvals, and learning rather than exporting a folder of files.
• The learning loop. Where customer permission, privacy rules, and contracts allow, aggregated or de-identified patterns can improve cold-start priors without exposing customer data. Platforms learn inside their own walls. Lapis can learn how advertiser decisions travel across them.
• The distribution loop. More channel coverage makes Lapis more useful to every advertiser. More represented demand makes deeper platform integrations and partnerships more valuable. Every new execution surface strengthens the reason to keep the advertiser objective in one place.

A new advertiser begins with priors. Model knowledge, forecasts, and competitor intelligence suggest the first tests. Permissioned customer outcomes validate or reject them. Conservative budgets, explicit uncertainty, and fast experiments govern the cold start. Then the private outcome memory begins to compound.

The result is three forms of compounding at once: customer-specific switching value, product learning, and broader distribution. A competitor can copy a generator. It cannot instantly recreate the accumulated state, permissions, integrations, and hypothesis-to-outcome history of the buyer. The base model can change underneath. The inventory mix can change outside. The advertiser relationship and record of decisions stay with Lapis.

Why This Becomes a Generational Business

A generational company needs more than a large market. It needs an entry point customers adopt, an expansion path that increases its value, an advantage that compounds, and a route to become infrastructure. Lapis has a structural path to all four.

1. The product sits upstream of $1.3 trillion in annual allocation. A creative tool competes for production budget. An independent buyer influences how media budget is allocated across the entire market.
2. The product expands with the customer. A business can enter through one brand and one campaign, then add products, users, experiments, channels, regions, and budget authority without replacing the system.
3. The value compounds with every campaign. Brand memory, permissions, hypotheses, outcomes, and operating history make the next decision better and the relationship deeper.
4. The distribution model reaches the whole market. Self-serve opens the relationship with founders and small teams. Team plans expand into growing companies. Managed operation serves businesses ready to delegate the loop. Agencies can use the same system across portfolios.

The economic ceiling follows decision authority, not the number of assets generated. Lapis can capture value without owning the media dollar. Self-serve plans monetize intelligence, planning, and production. Managed plans monetize execution and optimization. As agents absorb repeated execution, each operator can supervise more campaigns, more channels, and more budget. Lapis turns agency value into software economics because repeated decisions become product behavior instead of new headcount.

Distribution compounds too. The website-to-campaign workflow delivers useful output before a business assembles an agency or integrates every account. Self-serve acquisition opens the relationship. Managed operation expands it. Deeper official integrations make Lapis recurring infrastructure. A new advertising surface can eventually reach a base of represented advertisers through one partner, while an advertiser can test a new surface without hiring another specialty team.

This is the execution standard: customer outcomes, retained revenue, cross-channel expansion, and managed budget must grow faster than human headcount. When they do, Lapis becomes more than a powerful product. It becomes the default buyer for the long tail and the distribution layer through which new media markets reach demand.

Why Existing Stacks Leave the Buyer Open

The market already has generators, platform optimizers, agency AI tools, dashboards, and commerce agents. Each attacks one layer of the problem. Lapis connects the layers into one buyer that works for the advertiser. That is the difference between another useful feature and the control point for a new era.

Generators stop at production

A generator ends when the asset exists. Lapis begins there. The buyer knows why the asset should exist, which audience and moment it serves, where it should run, how much to spend, what happened next, and what to create after the result. Creative gives the system actions. The operating loop turns those actions into intelligence.

Platform agents stop at the auction boundary

OpenAI, Google, Meta, and Amazon will build powerful creation and optimization systems. Performance Max and Advantage+ already automate many decisions inside Google and Meta inventory. Local execution belongs to these sellers. Global allocation belongs to the advertiser’s representative. At full deployment, Lapis decides which seller deserves the budget, supplies stronger brand and creative inputs, reconciles results with permissioned first-party outcomes, and keeps the learning portable when spend moves.

Media owners control APIs, formats, eligibility, review, auctions, and delivery. Official interfaces lower friction while platform risk remains. Lapis preserves value when any one interface changes because customer-owned accounts, portable brand and experiment state, first-party measurement, channel-ready creative, and operating workflow survive the change. The auction can change without erasing the advertiser record.

DSPs execute a plan they do not own

A DSP receives a business objective after humans translate it into media settings, audiences, budgets, and finished creative. It optimizes execution after the central choices have been made, and it cannot turn every walled garden into one neutral auction. If the objective and creative arrive from upstream, the DSP remains an execution engine. Lapis can use DSPs while retaining the buyer role.

Marketing clouds are the closest enterprise threat

Marketing clouds, retail-media tools, and holding-company platforms already own integrations, data pipelines, workflow, first-party data, and enterprise distribution. Some are moving upstream into planning and creative. Lapis wins by owning richer advertiser context before the bid, productizing the full loop for customers their economics underserve, moving faster across emerging surfaces, and preserving learning across competing sellers.

Enterprise stacks sell collections of specialist tools to institutions. Lapis sells the advertiser institution as one product. That is the opening, and it is an execution race. The winner will be the company that turns context into reliable action fastest while earning the customer’s permission to do more.

Generic AI supplies a model, not a buyer

A foundation model can reason and generate. It does not arrive with durable customer permission, platform-specific integrations, outcome history, budget controls, policy state, or an audit trail. Lapis can replace the model underneath without losing the advertiser relationship, accumulated memory, or action surface. The model is an input. The governed loop is the product.

AI turns agency execution into software

The major agency groups will adopt AI, just as they adopted programmatic. Their strongest strategists, producers, and advisors will remain valuable. Their limitation is economic. Services scale through staffing, and operational memory remains fragmented across people, decks, files, and vendors. Lapis converts repeatable execution and memory into a persistent product. One operator can supervise work that once required a floor of specialists. Agencies can use and distribute Lapis, but they no longer gate access to the capability.

If a holding company builds the same architecture, it validates the category by becoming a software company itself. The minimum team required to access agency-grade capability still collapses, and the market opens to millions of businesses the holding-company model could never profitably serve.

The Necessary Constraint: Governed Autonomy

The future is a governed system with a constitution. The business defines what may be optimized and what may never be traded away before the buyer receives a corporate card and an instruction to “grow.”

Autonomy begins supervised. Agents propose and execute bounded work. Humans approve high-risk actions. Permission expands only after repeated evidence. The 2026 Stanford AI Index shows that enterprise agent deployment remained early in 2025. That makes constrained permissions, monitoring, auditability, and measured performance the adoption path. Reliability is earned one bounded action at a time.

Every buyer also works with imperfect attribution. A serious system connects permissioned first-party outcomes, runs controlled tests where possible, distinguishes correlation from incrementality, expresses uncertainty, and avoids overreacting to small samples. Persistent experimentation and memory beat a sequence of teams that forget context between campaigns.

The human role moves up the stack. People decide positioning, product truth, moral boundaries, taste, risk, and what the company is trying to become. The agent supplies speed, breadth, consistency, and relentless iteration. The best system concentrates human attention on the decisions that deserve it.

The Logical Conclusion

DoubleClick industrialized digital ad delivery. Right Media pioneered impression-level exchange trading. AdMob helped scale mobile-web and in-app monetization. Invite Media made cross-exchange bidding and optimization practical through a DSP. Admeld helped publishers compare demand sources and optimize yield. Those were foundational companies, but they acted after the central decision-maker had gone to work.

AI reaches the decision-maker itself. It can understand the business, hold the brand in memory, make the campaign, operate the tools, interpret the result, and improve the next attempt. That changes the scarce resource from access to competent judgment. It changes the product from software a media buyer uses to software that can perform much of media buying. And it changes the market from thousands of sophisticated advertisers supported by large institutions to millions of businesses with their own persistent advertising intelligence.

This is the AdSense-shaped opportunity. A forbidding market is becoming accessible at far lower fixed cost for the long tail. Context is becoming the matching language. The default on-ramp can become one of the most valuable layers in the system.

The company that owns the auction will be powerful. The company that owns the advertiser’s objective across auctions can be more enduring. It will carry the brand from Google to Meta to ChatGPT to the next interface, decide where the next dollar belongs, generate what each moment requires, and learn across boundaries no seller can cross. That is the software-native buyer control point the first ad-tech revolution never fully produced.

AdSense gave every publisher a sales force. Lapis gives every business a media buyer.

This is the company we are building at Lapis. The media platforms will own their auctions. Lapis will represent the business across them. The first ad-tech revolution could not build that company. AI makes it possible now.

Build With Lapis

The future starts with one campaign. Give Lapis your website and brand context, describe a business goal, and generate the strategy and on-brand variants that would otherwise require multiple specialists and handoffs. Use the self-serve product when your team wants control of planning and creation. Use a managed plan when you want the agentic loop operated across supported channels on your behalf.

Create your first campaign with Lapis, or see self-serve and managed plans. Lapis is independent from the media platforms it supports; campaign access, approval, placement, and delivery remain governed by each platform. That independence is the point. Lapis is building for the advertiser.

Related guides

• Agentic Ads: The Complete Guide: how the perceive, decide, act, learn loop works
• The Future of Ad Agencies and Ad Tech: what changes for holding companies and intermediaries
• Will OpenAI Build This?: the structural case for an independent layer
• LLM Ad Infrastructure Explained: the emerging stack beneath AI-native ads
• Can AI Replace an Ad Agency?: where humans remain essential