Paul Hewett, CEO of In Marketing We Trust is over in Texas for SXSW Austin. But rather than simply being on a jolly, B&T has set him to work as our on-ground correspondent. Here’s his second diary entry.
Monday morning started with Steven Bartlett, host of Diary of a CEO, on stage at 9 am in the Innovation Clubhouse before the main programme had even kicked in. The room was intimate, the setting was conversational, and it turned out to be one of the best sessions of the week. He was paired with CJ Bangah, PwC’s Front Office Transformation partner, for the Daily Decode, a recap format where people who have been absorbing the conference all week build on each other’s observations rather than repeat their own talking points. The chemistry between them worked. Bangah brought the analytical frame, Bartlett brought the pattern recognition and the willingness to say things out loud that most people only say in private.
Bangah’s read on the week’s evolution was useful. SXSW 2024 was all optimism, AI as a collaborator, the bright shiny object. 2025 brought a harder edge: misinformation, algorithmic control, digital hate. This year the tone has matured again. People aren’t asking what AI can do anymore. They are asking what it should do, who decides, and what happens to the people in the middle. The metaphors surfacing across sessions captured the mood. Bangah described AI as a kindergartener that we are treating like an adult. And a line from the week that stuck with me, also via Bangah: you can take a forklift to the gym, but that is not the point. AI will get you outcomes, but efficiency for its own sake is not the goal. For marketers who have spent the last 18 months being told to “do more with less,” that distinction matters.
Bartlett built on that with something more concrete. He introduced a phrase worth taking home: “obvious pain paralysis.” The pain of making a transformation decision is immediate and visible (we are going to have to redesign this entire org chart, we are going to lose people). The upside is vague and distant. You cannot quite quantify it. He made the point that what you hear on panels and what you hear from CEOs in private are two very different things. The public narrative is optimism and opportunity. The private conversations are closer to what one CTO described to him as steering the Titanic with a paddle. Anyone who has tried to get AI adoption past a leadership team that is simultaneously excited and terrified will recognise that description.
The Klarna story came up, and Bartlett addressed it directly. He said he had already heard the widely circulated narrative at the event earlier in the week (that Klarna replaced hundreds of staff with AI agents and had to reverse course). He called it out as not true. He said he spoke directly to Klarna’s CEO. The company had roughly 6,000 people last year. By the end of summer, it was around 5,000. The direction of travel continues, but the savings are being reinvested into human-centric work to improve customer experience. They are building a bigger company with fewer people. Later the same day, Ian Beacraft told the alternative version in his session: Klarna let go of 700 people, replaced them with AI agents, then had to rehire when complaints surged. Two credible speakers, same day, contradictory accounts. That gap tells you something about how poorly understood this transition still is, even among the people closest to it. It also tells you to be sceptical of any case study being used to sell you an AI platform.
Bartlett also described a 60-day AI transformation challenge he ran inside Diary of a CEO. The winning team built an agent that automated guest research and pitch preparation, saving roughly 72 hours a week. The agent researches the guest, analyses YouTube audience preferences, and compiles segmented pitch reports automatically. But the metric he cared about was not just efficiency. It was employee engagement. The framing was not “your job is at risk,” it was “your job is about to get more interesting.” He was honest enough to add that he does not see that level of enthusiasm at every business. And he pushed into territory that should matter to anyone managing a team right now: work has become one of our primary sources of community. As personal lives increasingly move to screens, the workplace is where real human connection is happening. If you hollow out the work, you are not just removing tasks. You are removing a source of belonging. That is not a technology problem. It is a leadership problem, and most organisations are not treating it as one.
Ian Beacraft – How to Design a Company That AI Can’t Outpace
The standout session of the afternoon was Ian Beacraft’s “How to Design a Company That AI Can’t Outpace.” Beacraft is the CEO of Signal and Cipher, and his argument cut through because it ignored the tooling conversation entirely. His core claim: execution has become cheap, but most organisations are still structured around the assumption that execution is expensive. Every approval layer, every departmental silo, every sequential workflow exists because humans were the limiting factor for 150 years. AI changes those constraints fundamentally, but if you just layer AI on top of the existing structure, you accelerate the failure rather than fix it. He drew a distinction that agency leaders in particular need to sit with: there is a difference between having 100 agents and agents being how work gets done. The first is sprinkling AI pixie dust on a broken system. The second requires re-engineering the system entirely.
The real shift, he argued, is from doing work to designing work. He laid out a three-tier hierarchy: operators (doing the work, 95% of what most people do today), designers (building workflows and systems that solve entire classes of problems), and architects (encoding organisational values, judgement, and culture into systems). His team spent two and a half years building what he calls AIOS (Architected Information Organisation Systems), which captures an organisation’s identity, decision architecture, governance, and quality standards into data sets that agents operate within. The practical output: agents that act to the same standard every time, not the same way every time. A six-person company called StrongDM is using exactly this approach to compete with organisations many times its size, with no human writing or reviewing code.
He was blunt about business models. The billable-per-seat model is dead. The hourly charge-out rate is dying. If you run an agency or any service business that charges for time, this is not a future problem. It is happening now. The economics of execution have collapsed, which means the value has to come from somewhere else: from the design of the system, the quality of the thinking, the strategic architecture. Not from how many hours someone sat in a chair. His advice on tooling was the clearest line of the day: if you focus on the tools, you will never recover from the whiplash. Focus on values, governance, quality standards. Build the foundation. The tools are just endpoints.
Monks – 25 Minutes of AI
Monks brought a session that marketers specifically should pay attention to. Oliver Koelemij and his colleague ran their “25 Minutes of AI” format, a rapid-fire briefing structured around intelligence (research and knowledge bases), create (content and code generation), and orchestrate (agentic workflows). The tool name-dropping was extensive, but the part worth sitting with was their Metropolis platform: a fully operational AI content pipeline that produces brand-compliant imagery, video, and social assets through orchestrated agent workflows. They showed AI-generated car imagery for an automotive client that was indistinguishable from traditional production photography, meeting strict manufacturer specifications. Production costs cut by 60-70%. Go-to-market compressed from months to minutes.
Think about what that means in practice. If a brand can produce high-quality content in near real-time, it can respond to cultural moments as they happen. Not next week. Not after a brief cycle and three rounds of revisions. Now. Monks used the phrase “move at the speed of culture” several times, and for once it did not feel like marketing language. It felt like an operational reality. For any brand or agency still running traditional production timelines, this is the competitive gap that is opening up. The question is not whether AI-generated content is good enough. Used well, it already is. The question is whether your production model can keep pace with the brands that have already made the switch.
Monks also landed on the same theme as Beacraft: the shift from “human in the loop” to “human in the lead.” Creators becoming curators. The role moving from producing individual assets to designing the underlying systems that produce hundreds of them. If you are only learning interfaces, they argued, you are already behind.
Between sessions, the hallways are quieter than in previous years and session quality remains wildly inconsistent. I went from the most insightful session of the week to the least insightful within minutes. You have to choose wisely at SXSW or you can waste a valuable hour.
But the through-line from Monday is hard to ignore. Different speakers, different rooms, different frameworks, all arriving at the same conclusion about a structural shift that is no longer theoretical. Something is building here.
More tomorrow.
