On 1 July 2026, Gartner put a number on a problem most Australian CIOs had already started to feel. Roughly US$234 billion of enterprise application software spend, close to a fifth of the total global SaaS market by 2030, is now exposed to what Gartner is calling agentic arbitrage. Nine days later, most FY27 SaaS renewals in this country are being reviewed against contract templates that predate the term.
That is the gap this piece is about. FY27 procurement calendars locked between May and June assumed the software you buy in July is priced, licensed and audited the same way it was priced, licensed and audited in FY26. It is not. The vendors know it. Some of your suppliers are already rewriting their own commercial models in response. The question is whether your renewal terms move with them, or whether you sign the same document you signed last year and discover the mismatch when the first agent goes into production.
What agentic arbitrage actually is
Agentic arbitrage is the term Gartner has settled on for a specific commercial effect. When an AI agent does work across two or more SaaS platforms on behalf of a person, that person does not need to open, click through, or hold a seat inside each of those platforms. The agent does the click. The agent reads the record. The agent updates the field. The human sees the outcome.
The vendor's per-seat revenue is built on the opposite assumption, which is that every human worker who benefits from the platform is inside the platform, holding a licence, generating a monthly fee. Break that assumption and you break the pricing model. You do not need every function to be automated for the effect to bite. Gartner's own modelling puts task-specific agents inside 40 percent of enterprise applications by 2026, up from under 5 percent in 2025. The commercial repricing follows the agent rollout by six to twelve months, which is exactly the window an FY27 renewal sits inside.
For an Australian enterprise, the practical implication is that your FY27 SaaS spend is being priced against a workforce that is about to stop touching the software the way it used to. If your renewal locks a per-seat rate for two or three years, you are pre-paying for licences your workforce will stop consuming. If your renewal locks an unlimited-use rate, you are absorbing the risk that vendors will re-price mid-term the moment agent traffic becomes visible on their side.
Why the July 1 Gartner number is the one to plan against
Analyst forecasts are usually easy to ignore in the year they land. This one is not, because it landed on the same week Microsoft launched Frontier Company, the same month Salesforce's Agentforce 360 became the default sales interface at ANZ, and the same quarter Wesfarmers stood up Gemini Enterprise agents across Kmart, Officeworks, Priceline and OnePass. The macro forecast and the local implementation timeline have converged.
The US$234 billion number is not a prediction of who wins. It is a measurement of contested revenue, meaning revenue that vendors will fight to protect and buyers will attempt to redirect. That fight is happening inside the renewal document. You cannot renegotiate a signed contract. You can shape a renewal.
The Australian angle procurement teams keep missing
Two structural features of the Australian SaaS market make the agentic arbitrage effect land harder here than in most jurisdictions.
The first is licence density. Australian enterprises consistently over-index on per-seat SaaS by global standards, partly because of the Microsoft anchor across public and private sector fleets, partly because Salesforce, Workday, ServiceNow and Atlassian all landed early and deep in the ASX 200. When Gartner talks about SaaS spending at risk, more of the Australian software budget sits inside the exposed category than in a comparable US or UK enterprise.
The second is contract cadence. A large share of Australian enterprise SaaS agreements roll on a fiscal-year cycle that ends 30 June, which means renewal negotiations happen in April, May and June, and signed paper lands in July. FY27 is now live. The paper being signed this week and next commits Australian buyers to pricing and permission terms drafted before agentic arbitrage was a named phenomenon.
Third, and this is where the risk compounds, most Australian SaaS contracts include an auto-renewal clause pegged to CPI or vendor price index. When your vendor re-prices agent traffic in FY28 by moving from per-seat to per-outcome or per-token, your CPI clause does not protect you. It protects the vendor.
The clause that needs to be in every FY27 SaaS renewal
If your organisation is going to run agents against a SaaS platform, and by 30 June 2027 most Australian enterprises will be, the renewal document should carry an explicit set of provisions on agent access. There is no single standard clause yet, and there will not be one for at least eighteen months. What the renewal needs is a bundle of positions, negotiated together, that answers seven questions.
One. Does an AI agent count as a user?
The default vendor position, unless you challenge it, is that any programmatic entity that authenticates and takes action inside the platform holds either a paid seat or falls under a separate machine-user rate. Salesforce, Microsoft, ServiceNow and Workday have all begun defining machine-user categories in their master agreements. If your renewal is silent, you inherit the vendor's definition, and the vendor's definition will move over the life of the contract. State, in the renewal, that agents acting on behalf of a licensed human user do not require additional per-seat licensing for that user's scope of work. Then define what "on behalf of" means, because that phrase is where the money sits.
Two. Whose agent is allowed?
Vendor-native agents are convenient. They are also the mechanism through which vendors will attempt to hold the customer inside a single stack. Preserve the right, in writing, to deploy third-party agents (your own build, an OpenAI agent, an Anthropic agent, a partner's agent) against the platform with the same data access, the same API rate, and the same authentication support as the vendor's own agent. If you do not preserve this right, your platform ownership becomes de facto exclusive over the term of the renewal.
Three. How is the agent authenticated, and what does that authentication cost?
Some vendors already gate agent authentication behind an enterprise API tier priced separately from the main seat licence. Others price at the OAuth token level. Others still are moving to a hybrid where a base API allowance is included but a per-call fee applies above a threshold. In every case, the price and the mechanism should be inside the renewal. If the paperwork does not name the cost, the cost is uncapped.
Four. What is the rate limit, and what happens when the agent hits it?
Human users rarely trip rate limits. Agents trip them on day one. Vendor policy on rate-limit breaches varies widely, from soft throttling through to account suspension. Ask the vendor to write into the renewal (a) the rate limit that applies to authenticated agent activity, (b) the notice period before throttling, and (c) the escalation path. If a Wesfarmers-scale customer service agent gets throttled during peak trading, you want the runbook already documented.
Five. Whose logs are they, and can you read them?
Agent workflows produce audit trails the vendor sees before you do. If a compliance question arises (an ASIC surveillance query, an OAIC privacy determination, an internal AFCA complaint) you need forensic access to the agent's activity inside the platform. Ask for read access to the vendor-side agent audit log, delivered in a defined format, on a defined timeline, at a defined cost. The Australian Cyber Security Centre's June 2026 guidance on agentic AI treats over-privileged agents and weak identity controls as the two most likely paths to compromise, and the audit log is the only evidence base that lets you investigate either.
Six. How does the vendor price mid-term repricing?
The vendor is going to want to reprice for agent volume at some point during the contract. Write in an option, not an obligation. Give the vendor the right to propose a new pricing model at defined intervals. Give yourself the right to reject it and revert to the base rate for the balance of the term. Do not accept a unilateral CPI-linked repricing clause on agent traffic.
Seven. What is the exit posture?
Agent orchestration builds vendor lock-in faster than any prior generation of enterprise software, because agents accumulate context, memory and workflow logic that does not port cleanly. Include, in the renewal, a defined export format for agent configuration, workflow logic, permission scopes and audit history. Twelve-month notice periods on exit are now table stakes. Ask for a data escrow clause covering agent state.
What this looks like inside a signed contract
The seven positions above do not add pages to a renewal. They add one appendix, sometimes called an Agent Access Schedule, sometimes folded into the Data Processing Addendum. Australian buyers with more than A$5 million in annual SaaS spend on a single vendor already carry enough weight to insist on it. Buyers below that threshold should form buying consortia through their industry body, or lean on their systems integrator to carry the terms on their behalf.
The ones we have seen close cleanly at Wai over the last quarter share three features. They name the AI agent as an entity in the definitions section. They price agent access as a separate schedule with its own volume brackets. And they carry a mid-term review provision that lets both sides adjust once real usage data is on the table, without letting the vendor reprice unilaterally.
The CFO reading
The line item on the FY27 P&L that a CFO should watch is not the SaaS licence line. It is the difference between what the licence line looks like at signing and what the effective per-user cost of the platform looks like six months after agents are in production. If the licence spend stays flat while the agent traffic scales, the vendor is absorbing the arbitrage. If the licence spend rises because of add-on API fees, machine-user licences, or throttling remediation, the arbitrage is landing on your P&L.
CFOs who have modelled this properly are now setting FY27 SaaS budgets against three usage scenarios, not one. A base case, in which agent adoption follows the vendor's own projection. A midpoint, in which internal deployment moves twice as fast. And a compressed case, in which agentic arbitrage forces a mid-term contract reopening. Australian enterprises that stress-test the P&L against all three, and then negotiate the renewal to fit the compressed case, are the ones that will not need to reopen the contract in FY28.
The board reading
The board question is simpler. If a fifth of the enterprise's software spend is exposed to a commercial repricing the vendor cannot fully control, what is management's plan for the terms of that spend, and where is that plan documented? Boards should be asking to see the FY27 SaaS renewal calendar, the list of contracts that touch agent-relevant vendors, and evidence that the seven positions above have been either taken or actively declined for a stated reason.
This is not a technology question. It is a procurement, finance and legal question with technology stakes. Directors sitting on ASX 200 audit and risk committees will want minutes that show the question was put to management before 30 June 2027.
What Wai is seeing across the market
Wai's ARC platform, which we run as the authority and AI-visibility layer for Australian technology and SaaS operators, tracks how enterprise buyers are moving on these positions in real time. Over the last eight weeks, Australian buyers we work with have shifted from adding agent language to their renewal templates as an afterthought, to leading the renewal negotiation with the Agent Access Schedule and letting the seat count fall out of it.
The vendor response has been surprisingly cooperative, mostly because the vendors that resist most publicly are also the ones most exposed. A vendor that refuses to define agent access in writing is a vendor that intends to reprice it unilaterally. That is a signal a procurement team can act on before signing.
The FAQ, answered directly
What is agentic arbitrage? Agentic arbitrage is the effect on SaaS pricing when AI agents perform work across multiple SaaS platforms on behalf of human users, reducing the number of humans who need to hold seats inside those platforms. Gartner puts the exposed SaaS spend at US$234 billion by 2030, roughly 20 percent of the enterprise application software market.
Do AI agents count as users under per-seat SaaS pricing? The default vendor position, unless negotiated otherwise, is that any authenticated programmatic entity taking action inside the platform is either a paid seat or a machine user subject to a separate rate. Salesforce, Microsoft, ServiceNow and Workday have all begun defining machine-user categories. Silence in the contract favours the vendor.
How should Australian CIOs prepare SaaS contracts for agentic AI? Negotiate an Agent Access Schedule as part of the FY27 renewal, covering agent definition, third-party agent rights, authentication and pricing, rate limits, audit access, mid-term repricing rights, and exit format. Refuse unilateral CPI-linked repricing on agent traffic.
Does an AI agent need a licence to use Salesforce or Microsoft 365? It depends on the master agreement version and the negotiated schedule. Under current default terms, agents authenticated against these platforms typically require either an incremental licence or an API-tier arrangement. Both vendors will negotiate on this in mid-market and enterprise renewals through FY27.
What happens to per-seat SaaS pricing when AI agents replace users? Per-seat revenue drops for the vendor while workload increases, which is why vendors are moving to hybrid pricing that combines a base seat rate with per-agent-call, per-token or outcome-based fees. Buyers that lock a pure seat rate in FY27 will face repricing pressure by FY28.
How do we negotiate agent permissions with SaaS vendors? Treat the Agent Access Schedule as the primary negotiation, not an afterthought. Bring specific asks on definition, third-party agent parity, authentication cost, rate limits, audit access, repricing rights and exit. Bring the compressed usage scenario, not the base case, as the pricing anchor.
Should we renegotiate SaaS contracts before deploying AI agents? Yes, particularly on contracts that renew during FY27. Waiting until after deployment gives the vendor the usage data first, and the vendor will use it. Renegotiate before agents enter production or align renewal timing with your agent rollout plan.
What is the Gartner $234 billion agentic AI number? Gartner's July 2026 estimate that roughly US$234 billion of enterprise application software spend will be exposed to agentic arbitrage between now and 2030, representing about 20 percent of the enterprise SaaS market. Best-case scenarios have agentic AI driving 30 percent of enterprise application software revenue by 2035, surpassing US$450 billion.
The last word
You cannot buy your way out of agentic arbitrage. You can only price it, permission it and audit it inside the contract before the vendor prices, permissions and audits it for you. FY27 is the window. The renewal is the mechanism. The Agent Access Schedule is the document.
Australian enterprises that get this right will not save 20 percent on their SaaS bill. They will retain the freedom to deploy the agents they want, on the platforms they hold, at a cost they can forecast, without a reopening clause hanging over the CFO. That is the outcome worth spending a week of legal and procurement time on this month.
Wai advises Australian technology, SaaS and enterprise operators on AI strategy, SaaS commercial design, contract terms for agent access, and AI discoverability inside answer engines through the ARC platform. If your FY27 SaaS renewals are still open, this is the four weeks that matter.