Quanta Bits - July 18, 2026

Hi everyone,

Greetings from smoky Boston. Smoke from wildfires in Canada and northern Minnesota drifted over New England earlier this week, gave us a couple of pretty bad days, and looks ready to visit again.

Most of my attention this week has been on how companies line up their technical people behind business priorities. AI is pressuring teams to move faster and work through more requests across organizations. Many technology organizations still split understanding the business problem, building the solution, and owning the applications among different teams. This week's essay is about the operating model behind that work, in particular how initiatives should be handled by technology teams, where technical resources sit, and who stays with a business problem until it works.

The Wall Street Journal reported on July 17 that workplace transcription is moving beyond scheduled calls. People are using phones, small wearable recorders, and apps such as Granola to capture office conversations, conferences, and other moments that never had a meeting link. According to the Journal, Granola had 758,000 mobile downloads and 122,000 mobile users in the second quarter of 2026, and it can transcribe without adding a visible participant or playing an automatic alert.

Recording these conversations can improve recall, but it creates immediate questions about consent, ownership, access, and deletion. Companies need an explicit policy before searchable company memory becomes the default.

The Financial Times reported on July 13 that DoorDash, Siemens, Airbnb, and Lindy are moving some work to lower-cost Chinese AI models. DoorDash sends simpler work to Kimi, a lower-cost model from Chinese AI company Moonshot AI, and reserves a more capable model for harder tasks. The Wall Street Journal found the other side of the market: Shopify requires frontier models, the most capable and often most expensive models available, because mistakes cost more engineering time. Spotify has declined to buy some expensive upgrades.

Choosing a model is becoming a workload decision. The right choice depends on the required quality, the data involved, response time, the cost of a mistake, and price. Companies need those rules before individual teams turn model selection into an unplanned spending policy.

Aaron Levie, the CEO of enterprise content-management company Box, shared notes on July 8 after meeting with a couple dozen enterprise IT leaders about AI agents, software that can take action across business systems. Agents work best when tied to a business process, he wrote, but those processes cross company silos. Measuring results also requires getting close to each workflow, and people who know how to implement agents inside companies are hard to find.

Many CIO teams still separate understanding the problem from building the solution. A business applications analyst talks with teams such as HR or Sales, writes requirements, and passes them to a developer aligned with a particular system. That model made sense when many requests ended inside one application and software was expensive to produce.

AI can now speed more than prototypes. With coding agents, shared components, and approved APIs, one capable person can build and ship smaller cross-system applications that once required several handoffs. Integration, permissions, testing, and adoption still take time, but the analyst-to-developer relay is not always necessary.

I think CIO organizations need a blended delivery lead: a domain expert with the AI-assisted development skills to build, ship, and improve solutions across the workflow using approved APIs, company knowledge sources, and shared components. This is not a requirements or coordination role. They need real development capability, though not the depth of a platform engineer. They decide whether work needs ordinary automation, an AI-assisted workflow, or an agent; build it within approved patterns; and bring in specialists for security, scale, or difficult integrations. They also need the judgment to decide whether the work is worth doing.

Product companies have traditionally divided the work between product managers and engineers. Product managers talk with customers, decide which problems matter, set priorities, and translate that context into requirements. Engineers make the technical decisions and turn those requirements into working software. The division made sense when engineering capacity was scarce and producing code took most of the time.

In a June 27 VentureBeat essay, Amazon software engineer Ishan Gupta argued that coding agents change that balance. When engineers can produce code much faster, deciding what should be built becomes a larger part of their job. Engineers need more customer and product judgment. Product managers need to contribute more than requirements, backlog management, and coordination. Their value moves toward understanding the customer, making harder tradeoffs, shaping adoption, and staying accountable for whether the product works.

Corporate technology teams have a similar division. Analysts understand the business process and translate it into requirements; developers understand the systems and implement what reaches their queue. AI-assisted development lets more people combine those skills. Some analysts can build the solution themselves, while some developers can work directly with the business and make informed product decisions. That is the shift behind the blended delivery lead: one person can stay with a problem from the first conversation through release and improvement.

Take employee onboarding. A request to improve it may touch the HR system, identity management, the intranet, training, service tickets, and payroll. An HR applications analyst can document the problem, but the work then moves through several technical teams. Every handoff adds queue time and another chance to solve only one piece.

An HR-aligned delivery lead would stay with the whole problem. HR would own the process goal and policy decisions. The lead would build and ship the workflow using approved APIs and company AI tools, bringing in identity, security, or platform specialists where needed. Together, they might track how quickly new hires get access, how many setup errors occur, and how many support tickets onboarding creates.

Applications increasingly expose data and actions through approved APIs, controlled ways for software to communicate. Some separate the user interface from the systems behind it, a design often called headless. Both support cross-system work without changing ownership. Application and data owners remain accountable for official records and access. Platform teams operate system connections and production services. The delivery lead uses those interfaces to build the complete workflow.

Delivery leads should remain inside the CIO organization and follow common intake, security, testing, release, and support processes. A central AI operations function can set shared standards for testing, monitoring, cost, and escalation. Application, data, and identity owners should still approve access to their systems.

An HR-aligned lead should know the roadmap and the people doing the work. A Sales-aligned lead should understand pipeline reviews and customer relationship management (CRM) data. A Marketing-aligned lead should understand campaign launches, lead routing, consent, and attribution. The business leader owns the process goal and adoption; the delivery lead owns building, releasing, measuring, and improving the solution. Success is whether the workflow improved, not how many tickets the technology team closed.

AI vendors already use versions of this model. Aaron Levie wrote in May that enterprise agent work will require vendor engineers and new internal agent engineering roles. On June 30, AWS announced a $1 billion group that puts engineers directly inside customer teams. These programs are a useful market signal, although an internal CIO model will look different.

CIOs can start without copying the vendor model or renaming the whole team. Pick one domain, one set of workflows, and one person with business knowledge and AI-assisted development skills. Give them an approved development environment, access to relevant APIs and company knowledge sources, clear decision rights, and specialist help when needed. Record the current result, then see whether the new model improves it.

The harder part will be developing the people. Some business application analysts will want to learn AI-assisted development well enough to build production applications. Some developers will want to learn a business domain well enough to make product decisions. Not everyone will, and that is fine. Deep platform specialists will still matter, especially when security, scale, or a difficult integration is involved.

Organizing most delivery around applications preserves a boundary the work no longer respects. As AI makes building faster and applications become less visible to users, CIO teams need more people who can build across the company's systems and stay with a business problem until it works.

Is there anything this man can't do? 😊

Source: djcows on X

Actor Eric Dane was losing his ability to speak as ALS progressed. In the final weeks of his life, ElevenLabs, a company that creates synthetic voices, recreated his natural voice from past recordings so he could communicate with his family. When his teenage daughters heard it, they said, "That's your voice." Dane died in February before the family could fully use it. ElevenLabs told Variety that its voice-restoration program has helped more than 7,000 people and pledged free lifetime licenses and support to people with terminal neurological diseases, with a goal of reaching one million. The case shows the potential of this technology when a person or family has clearly chosen the use.

Book: Shelby Van Pelt | 368 pages | ★★★★★
Film: dir. Olivia Newman | 111 minutes | ★★★★☆

I read Remarkably Bright Creatures a few years ago and loved it. You can mostly see where the story is going, and it has a little of the feel of a Disney movie, but it still works. Tova is looking for closure after losing her son, Cameron is searching for his family, and Marcellus the octopus understands the connection between them before any of the humans do.

The movie keeps what mattered to me: grief, family, closure, and the idea that what you are looking for may be right in front of you. I enjoyed it, and the ending still landed. But the book is the stronger version. It gives you more time with Tova, Cameron, and especially Marcellus, who matters much more on the page. The movie is worth watching. The book stayed with me.

Prefer to listen? Quanta Bits is also available on Apple Podcasts and Spotify.

I collaborate with Spock, my AI agent. He researches extensively: scanning, filtering, and surfacing what's relevant across my business. I read, listen, and watch what resonates, and decide what matters. I provide direction, we draft together. The editorial judgment is mine. He'd tell you the same. Most logical. 🖖