Field notes, August 2026: budgets, commerce agents, and the planning season
AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.
Agentic marketing in late summer turns from building to accounting. Planning season starts, and the questions we keep returning to are about evidence: what agent programs can show for this year, and what they will need to show next year.
These are our monthly notes: themes we are tracking, questions we are studying, and what we expect. They are working hypotheses from our research, not measurements of the market.
Planning season asks for evidence
August is when many organizations open next year's budget discussions. For agent programs, we expect this to be a moment of accountability. A pilot that felt useful in spring now has to compete for money against headcount, media, and other software.
The theme we are tracking is the gap between programs that measured their work and programs that did not. An agent can be helpful in ways everyone on the team feels and still be hard to fund, because "it saves us time" is not a line a finance partner can check. Programs that recorded a baseline before launch, and tracked the same metric after, have something to bring to the table.
Consider a hypothetical team that used an agent to draft product descriptions this year. If it logged hours per description before and after, plus the human edit rate, it can make a specific case. If it logged nothing, it can only describe a feeling. We think the first case will win more often, even when the second agent did equally good work.
The practical advice is to start measuring now, even late in the year. A few months of clean data beats none. Our 2027 budgeting guide describes how to build a case from what you already have.
Commerce teams start with product data
The second theme is how retail and commerce teams are thinking about AI shopping assistants and agent-initiated purchases ahead of the holidays. We do not know how quickly agent shopping will grow, and we are wary of anyone who claims to. What we can say is which preparation looks sensible under almost any scenario.
The first step we recommend is unglamorous: fix product data. Complete attributes, accurate prices and stock levels, and clearly stated policies help with search, marketplaces, and assistants alike. If agent shopping grows slowly, the work still pays off in existing channels. If it grows quickly, you are ready.
The gap we expect teams to find is in attributes that live in people's heads. Merchandisers often know which products suit which recipients, occasions, or skill levels, but that knowledge rarely reaches structured data. An agent can propose these attributes from existing descriptions, and a merchandiser can approve or correct them in batches.
A pattern worth testing is to treat the product feed as the primary surface and the product page as one rendering of it. That inverts how many teams work today, where the page is written first and the feed is an export. We explore the idea in Product Feeds Are the New Landing Pages.
Operating discipline as the differentiator
The third theme is the one we find most durable. As agent tooling matures, we expect the difference between strong and weak programs to come less from technology choices and more from how the program is run. Two teams with the same models and tools can get quite different results.
The operating habits we think matter most are easy to list and harder to keep up. We use it as a checklist when we review agent designs in our research.
- Named owners: each agent has one person accountable for its output.
- Weekly reviews: someone reads a sample of agent work on a fixed schedule.
- Versioned instructions: changes to prompts and rules are tracked and reversible.
- Test sets: a fixed set of examples is rerun before each change ships.
- Failure budgets: the team agrees in advance how much error is acceptable.
- Clear permissions: each agent can touch only the systems its job requires.
None of these items requires new technology. Each requires time and attention, which is why we expect them to separate programs more than model choice will.
These habits also connect back to the budget theme. A program with weekly reviews and a test set produces evidence as a side effect of running well. A program without them has to reconstruct evidence after the fact, which is slower and less convincing.
Sales and marketing alignment around agents
The fourth theme is the handoff between marketing and sales. When agents produce leads, research briefs, or account summaries for sales teams, both sides have to agree on what a good handoff looks like. That agreement can be the start of better collaboration.
It can also go the other way. A handoff that arrives early, with no explanation of why the lead qualified, can damage trust in a way that is slow to repair. Our working view is that transparency is the deciding factor: show the evidence behind each handoff, and give sales a simple way to say it was wrong. We cover the mechanics in Building Trust With Sales.
Questions we keep returning to
Several open questions came up repeatedly in our research this month. We list them because we have not settled them, and we would rather test them in the open.
- Budget structure: should agent programs sit in budgets as software, as headcount, or as their own line?
- Product attributes: which attributes matter most for being found and recommended by AI assistants?
- Speed as value: how should teams evaluate agents whose value is mostly speed rather than quality?
- Baseline timing: how much pre-launch data is enough to make a before-and-after comparison credible?
Our current leaning on the first question is that a separate line makes costs visible and easier to defend. That is a hypothesis, and the right answer probably depends on how each finance team already handles software and services.
What we are watching next month
September is when preparation for the fourth quarter becomes urgent. We will watch three areas closely and report what we learn.
- Year-end readiness: how programs test agents, tools, and review queues before year-end volume arrives.
- Consolidation: whether the Agentic Hub pattern gains ground as programs bring several agents under one coordination layer.
- Vendor claims: how fall conference season shapes what vendors say agents can do, and how specific those claims become.
The open question we are carrying into September is whether budget pressure will push programs toward fewer, better-measured agents, or toward more agents with thinner oversight. We expect the first, and we will be watching for signs of the second.
How to apply this: pick the one agent you most want funded next year and start logging its baseline metric this week. A few months of clean before-and-after data will make a stronger case than any description of how useful it feels.
Related research
Fall conference season: questions to ask agentic marketing vendors
Many booths this fall will promise agents. These questions help separate products ready for real marketing work from demos that only work on stage.
Field notes, July 2026: evaluation becomes a team sport
Our monthly notes on agentic marketing. This month: why we expect evaluation to move out of the technical corner and into the whole marketing team.
Field notes, September 2026: year-end readiness and the hub model
Our September notes on agentic marketing: why review capacity may decide year-end results, and why we expect agents to consolidate under one hub.
Put this research to work
Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.
