AIforce was framed as AI replacing the UI. A Salesforce president demonstrated a working interface he had built by talking to Claude over a few days. The pitch is that people will update records and trigger workflows without ever opening the Salesforce app.

Enterprise software is admitting that the screen it spent twenty five years perfecting is no longer where the work happens. Now apply that to what you actually make. If the customer journey increasingly runs through an agent that reads, summarizes and recommends, then the site, the landing page and the beautifully art directed product page are all being mediated by something you did not design and cannot see.

Nobody in the agency business has a good answer for this yet. Which means the question is worth more than the answer for the next two quarters.

So what. Add one slide to every credentials deck that asks the client where their brand shows up when the customer never opens the app. Bring a point of view, even a half formed one. The first agency in the room who raises it owns the conversation.

Commerce Cloud shipped a shopper agent. Benioff demonstrated it by describing a single conversation that produced ten complete outfits, with the agent checking inventory and shipping windows and generating backup options in case something fell through.

Sit with that for a second. The thing evaluating the product was not a person scrolling. It was a system weighing availability, delivery date, price and fit against a stated intent. It never saw the campaign. It never saw the packaging. It did not feel anything about the brand.

Every discipline in an agency is built around persuading a human with attention, emotion and time. A meaningful and growing share of purchase decisions is about to be filtered through something with none of those. That does not make brand worthless. It changes where brand has to do its work, which moves upstream to the moment a person forms the intent they hand to the agent, and downstream to the structured product data the agent can actually read.

So what. Two new questions for every commerce client. What makes a customer ask for us by name rather than by category. And is our product data clean, complete and machine readable enough to win when an agent is comparing us to four alternatives. Most brands fail the second one badly, and almost nobody is being paid to fix it.

Marketing Cloud got an agent that, in the coverage of the keynote, builds campaigns by itself. Benioff sold it with a joke about marketers hand building five hundred campaigns in the middle of the night, and declared that work finished.

Be honest about what that describes. Versioning, adaptation, channel resizing, audience splits, subject line variants, always on CRM flows. A large amount of agency revenue is exactly the work he just put on stage as a solved problem, and it is sitting inside the platform the client already licenses.

This is not a reason to panic. It is a reason to be extremely clear eyed about which of your scopes are volume based and which are judgment based, because the volume based ones are repricing whether you participate or not. The agencies that get hurt are the ones whose profitability quietly depends on production hours they have not defended in a decade.

So what. Run the numbers this month. What percentage of your revenue comes from scopes where the deliverable is quantity. Then decide whether you are going to compete on cost, which you will lose, or move that money into strategy, brand, governance and the things in takeaway ten. Do it before a client does the math for you.

Salesforce published its architecture five days before the keynote and named the thesis plainly. AI reasoning can be open ended, while enterprise execution often cannot be. The model layer, the cloud layer and the interface layer were all deliberately made into choices. The governed business context underneath was not.

This is the single most useful idea of the week for an agency. Your tools are swappable. Whatever AI stack you paid for last quarter will be cheaper or free next quarter, and your competitor down the street has the same one. What is not swappable is the specific, accumulated, hard won understanding you hold about a category, its customers, its regulatory edges and what has already failed in it.

Every indie I work with has this and undersells it, because it feels like table stakes from the inside.

So what. Cut the slide about your AI tooling. Replace it with proof of category depth. The years, the accounts, the pattern recognition, the three things you know about this business that the client's own team has not articulated. That slide does not depreciate.

The dedicated Agentforce keynote was almost entirely customers and numbers. Eighty thousand hours saved at Fulton Bank. Three hundred and eighty nine million dollars in loans and deposits attributed. Four days off complaint resolution. Thirty thousand customers live on the platform.

Two years ago a demo was enough. Now the biggest company in the category cannot get through a keynote without a scoreboard, because the market has stopped buying possibility. Your clients are in the same mood, and marketing has less measurement credibility than customer service does.

This hits indies harder than it should, because indies are often better at effectiveness than they are at documenting it. The work is there. The case study is not.

So what. Every case study in your deck needs a business number and a source. Not impressions, not engagement rate. Something a CFO would recognize. If you do not have one, call the former client and ask. Most of them will tell you, and the call itself is worth making.

Adecco piloted Agentforce Coworker in the UK and France, then announced a rollout across more than forty countries. It was held up all week as the model, and specifically as the opposite of the common failure pattern where a proof of concept on clean data gets declared a win and scaled sideways.

Indies have been told forever to chase the big integrated AOR engagement because that is where the margin and the security are. That advice is now working against you. Clients are being trained, publicly and at scale, to expect a small proof followed by an earned expansion. Project work stopped being the consolation prize.

The upside is real. A holding company cannot staff or price a tight ninety day engagement without losing money on it. You can.

So what. Build a productized ninety day engagement with a fixed fee, a named outcome and a decision gate at the end. Lead with it instead of the full capabilities pitch. You are not selling small, you are selling the shape they already trust.

The most concrete thing announced all week was seven agents with human names and job descriptions. Casey. Paige. Carter. Marshall. Piper. Fin. Hunter. Each one mapped to a role a buyer already staffs and budgets for. They got more traction in the coverage than the architecture underneath them.

Agencies chronically overpitch capability and underpitch the named, scoped, staffed thing. We say integrated strategic partnership when we should say a senior strategist, two days a week, who owns your category intelligence and sits in your Monday meeting. Specificity is not a downgrade from vision. It is what makes vision purchasable by a procurement team that needs a line item.

So what. Go through your offering list and give every service a name, a named owner and a description a procurement person could put on a purchase order without calling you first. This is boring work and it moves close rates.

Salesforce showed a long horizon agent runtime built as a cycle of goal, plan, human in the loop, signal and continuous refinement. Agents can work toward bigger goals over longer periods, while people keep the ability to step in, redirect and approve.

The most aggressive AI vendor in enterprise software just told the market that human judgment is a designed component of the system, not a workaround for when the machine fails. That is the permission structure agencies have been waiting for. You are allowed to charge for taste, judgment and craft again, out loud, without sounding like you are protecting your job.

The trap is claiming it vaguely. Human in the loop as a comfort phrase is worthless. Naming the specific moment where a person intervenes, and what changes because they did, is the whole argument.

So what. Put your people back on the org chart slide with names and faces, then annotate the process with exactly where a human decides something. Show the two versions, the one the system produced and the one your creative director changed. That comparison is the sale.

The sharpest practitioner commentary of the week argued that the refusal set is the part teams skip and the part that decides whether a deployment survives its first bad month. Not a vague instruction to be helpful and safe, but an enumerated list of what the system must never do, with a test proving each refusal fires. Separately, three of the most powerful people in AI took three different positions on pacing and regulation from the same stage in the same afternoon.

Every agency has an AI policy slide. Almost all of them are aspirational mush about augmenting human creativity. Almost none of them say what the shop will not do, which is the only part anyone remembers.

In this industry the lines are obvious once you write them. No synthetic likeness of a real person without written consent. No AI generated talent standing in for talent you would otherwise book. No model trained on a client's proprietary assets without an agreement that says so. No unlabeled synthetic imagery in earned or editorial. Your list will be different. It should cost you something.

So what. Write it this month. Four to eight lines, agreed by leadership in one meeting. Put it in the pitch deck. It will be the most memorable page in the room, because every other page is a promise about what you can do and this is the only one that is a promise about what you will give up.

The consistent practitioner warning all week was that announcements arrive at vendor pace while adoption happens at the pace the data and permissions allow, and those two speeds have never matched. The recommended first move was an audit, not a platform purchase.

Here is the agency version, and it is the biggest commercial opportunity in this report. When a client's systems start generating campaigns, answering customers and recommending products on behalf of the brand, something has to define the layer those systems speak from. Voice and tone with actual examples. Claims that legal has cleared and the exact wording. Product truths. Category vocabulary. Competitor comparisons you will and will not make. The things the brand refuses to say.

Right now that work is being done badly, by nobody in particular, in a shared drive, by whoever was in the room. It is brand governance with a new and much more consequential job, and it is recurring by nature because it has to be maintained.

So what. Name it, scope it, price it as a retainer and take it to your three best clients this quarter. It is the most defensible recurring revenue available to an independent agency right now, and the management consultancies have not noticed it yet.

Salesforce's applications president told the room that the company's real product is the trust customers place in it to hold their data, workflows and security rules. Benioff spent a full segment insisting customer data does not go into the models. That was positioning, not housekeeping.

When capability becomes commodity, trust becomes the differentiator. That is the entire thesis of Relationship Selling, arriving now as the explicit strategy of one of the largest software companies on earth.

Independent agencies have a structural advantage here and mostly waste it by talking about process. The people who pitch are the people who do the work. The founder answers the phone. Nobody gets swapped out ninety days after the win. Those are trust claims and they are the ones holding companies cannot make.

So what. Make the continuity claim early, make it specific with names, and be willing to write it into the contract. A staffing guarantee with teeth is worth more than any capability slide you own.

Four of the most consequential announcements landed in the weeks before the keynote. The architecture memo went out five days early. By the time forty three thousand people were in the theatre, the audience was not learning the thesis, it was confirming it.

Agencies still treat the final presentation as a reveal. We hoard the thinking for ninety minutes and hope a room of eight people, most of whom have never met us, assembles it correctly in real time while also judging our chemistry. It is a strange bet and we keep making it.

The people who decide are rarely all in the room, and the ones who are will have to re sell your idea internally without you there. Give them a head start.

So what. Put the organizing idea in front of the client before the pitch. A short memo, a call, one provocation they did not have. You are not giving away the thinking, you are making sure it survives contact with people who never saw the deck. Then the room becomes the moment they agree instead of the moment they decide.

Forty three thousand people. One thousand four hundred and thirty published sessions. The most talked about moments of the week were two CEOs wandering through the crowd riffing at each other, and a cartoon of a co founder popping up in the corner of a demo. Elsewhere on the program, a session named attention as the scarcest resource in business.

Salesforce spent a fortune on the largest software event on earth, and what cut through was human, unscripted and slightly ridiculous. That is not an accident and it is not a criticism. It is the founding fact of our industry showing up inside somebody else's conference.

Every agency knows this about advertising. Very few apply it to their own pitch, where we rehearse the spontaneity out of ourselves and then wonder why four shops felt the same.

So what. Look at your next final presentation and ask what in those ninety minutes is genuinely unrepeatable by anyone else in the process. If the honest answer is nothing, you have a deck rather than a pitch, and you are going to lose to whoever shows up as a human being.