What are the best automation tools for an agency? Increasingly, the ones Meta and Google already hand you. The platforms are automating targeting, ad building, keyword work and campaign setup themselves. A sound agency AI strategy stops trying to out-build them, buys what it needs off the shelf and puts its own effort where client judgement is the product.
Even the biggest networks are finding out how expensive building is. Adweek reported in August that Omnicom moved at least 468 of the product and engineering staff behind its Omni AI platform, across the US, UK, India and Malaysia, to IT services firm Endava in June and July. That followed about 50 US layoffs in the Omni Platforms Product & Engineering division on 9 June, and it happened while Omnicom was pitching the overhauled Omni platform as the keystone of its post-IPG client offering.
If a holding company with Omnicom's resources hands the people who build its platform to a contractor, a 40-person independent should think hard before hiring its first AI engineer to rebuild Ads Manager.
Which agency jobs have the platforms already automated?
One Singapore agency founder, who spent 18 months trying to make his own shop AI-native, showed a slide at an AI adoption session this week that deserves a place above every media director's desk. As the diagram shows, it maps four jobs agencies used to hire people for against what the platforms now do instead:
- Meta audience stacking: Advantage+ treats your targeting as a suggestion.
- Meta building every ad: Meta aims to automate ad creation end to end.
- Google keyword lists: Dynamic Search Ads move to AI Max.
- Google campaign setup: an assistant sets up campaigns from a plain-English request, with your approval.
His conclusion was that agencies should not build tools for SEO, Meta or ads management that the platforms will do for them. That is an old race, and the platforms own the track.
For a performance or paid social shop, this is uncomfortable reading. A large chunk of what junior media executives were billed for is turning into a setting. The work does not vanish. It moves upstream, to briefing the machine, checking its choices and explaining the results to the client.
Why is building your own AI tools an old race?
Because the build is the cheap part. The founder's rule was "buy, never build", and his reasoning came down to one question: who is going to maintain it? APIs change, models change, platform policies change. Every internal tool becomes a permanent line on the payroll.
He also drew a distinction worth keeping: these are tools, not solutions. A tool automates a step. A solution changes an outcome the client cares about. An agency that builds a bespoke bid optimiser has built a tool, and is now competing with the platform's own optimiser, which has more data, more engineers and every incentive to win.
The Omnicom story is the large-scale version of the same lesson. Building a platform is a decision to run a software company alongside your agency, indefinitely.
Why does this keep happening to agencies?
Michael Porter's Five Forces is a simple way to see it. The model says the profit in any business is shaped by five pressures: rivals, new entrants, buyers, suppliers and substitutes. Two of those are pressing on agencies at once.
The platforms are powerful suppliers: they control the inventory, the data and now the tooling. They are also a growing substitute: every task they automate is one a client no longer needs an agency to do. When your supplier becomes your substitute, competing with it on the same task is a losing position.
The instinctive response is to keep doing what worked, only faster. That is the quiet trap. Stagnation is rarely dramatic. It looks like a team getting very good at audience stacking in the year it stops mattering. The opportunity goes to the agencies that move their effort to where the platforms cannot follow.
Where should agency AI effort go instead?
Into the work where your expertise and your client judgement are the product. The platforms optimise inside their own walls. They do not tell a client which tension in the category is unclaimed, how a rival is repositioning across every channel, or which idea is worth the risk. That is the agency's job, and AI can make it faster without making it generic.
The market is drifting the same way. In Digiday's Media Agency Report 2026, 77% of agency respondents said they plan to raise budgets for AI search and GEO strategies over the next 12 months, while 36% said client media budgets were flat this year. Flat budgets and new channels mean clients need advice, not more button-pushing. Even agentic buying, the next big automation promise, remains unsettled: "Agentic buying is a big topic, but it's a pretty ill-defined one right now," Tinuiti's Harry Browne said in the report.
What that looks like by agency type:
- Media agencies: stop selling setup hours; sell the plan, the measurement logic and the guardrails around platform automation.
- Creative agencies: let the platform generate variants; own the idea, the brand standards and the call on what ships.
- CRM and influencer agencies: automate reporting and monitoring; spend the time on audience insight and creator selection.
- PR and social agencies: automate the listening and the clippings; keep the narrative and the crisis calls human.
How do you decide what to build, buy or leave to the platforms?
Run every proposed AI project through a simple filter before anyone writes code or signs a contract.
| Type of work | Examples | Who is automating it | Your move |
|---|---|---|---|
| Platform execution | Audience targeting, ad variants, keyword lists, campaign setup | Meta and Google | Use the native automation; own the brief, approvals and QA |
| Generic productivity | Meeting notes, first drafts, resizing | Off-the-shelf AI tools | Buy; set standards for how they are used |
| Category and audience intelligence | Competitor ads, conversation, personas across platforms | No single platform does this for you | Buy a specialist stack; do the interpretation yourself |
| Strategy and creative judgement | The idea, the positioning, the advice | Nobody, and nor should they | Put your people's time here |
Then turn the filter into a plan:
- List every recurring task your team bills or absorbs. Reports, audits, setup, resizing, timesheets. Be honest about the hours.
- Mark the ones a platform is already automating. Use the four rows on the slide as a starting point, then check your own channels.
- Stop investing in those. Move the people to supervising the automation and explaining the outcomes.
- Buy, don't build, for the generic work. Ask every vendor who maintains it and how you get your data out.
- Name the two or three areas where your judgement is the product. That is your strategy, and where AI should make your people faster and better informed.
- Review the list every quarter. The platforms ship faster than your planning cycle.
What does this mean for independents versus the networks?
Better news than it sounds. The networks spent heavily on proprietary platforms and are now, as the Omnicom move shows, rethinking who builds and maintains them. Independents never had that sunk cost. They can buy capability as a service and move faster.
That is the thinking behind the Defy platform: the same analytical and AI capability as the holding networks, without the build cost or the maintenance burden. The platforms will keep automating execution. Your edge is knowing what to tell the client to do about it.
