Darrell Ross @darrellross

“What allows an organization to safely delegate action?” That’s the primitive underneath all of this. When you read the Claude Tag description, it’s actually describing four new organizational capabilities: 1.Shared memory 2.Context accumulation 3.Initiative 4.Asynchronous execution Those aren’t software features. They’re organizational design behaviors. The moment Claude can: •remember prior work, •observe multiple conversations, •take initiative, •continue working without being explicitly instructed, you have effectively created a new organizational actor. Which means the enterprise suddenly has to answer questions it previously avoided: •What decisions can this actor make? •What authority does it have? •What information can it access? •How is it evaluated? •Who is accountable for its actions? That’s why I think @emollick is correct that AI adoption is becoming an organizational design problem.

Ethan Mollick @emollick

5 days ago

Decisions about how to use AI in your organization are increasingly organizational design and strategy decisions, not IT choices: How do you integrate agents into your firm? What intelligence will you outsource? What are the boundaries of the firm? What is the role of people?

52 79 639 76K 406

The deep dive into this topic: Air-conditioning America: Engineers and the Controlled Environment, 1900-1960 (Johns Hopkins Studies in the History of Technology) a.co/d/0hxRzUlh Fascinating history of the engineers behind early HVAC. You realize that HVAC has been more impactful than most innovations (and yes, more impactful than the Internet) Commercial air travel, sky scrapers, enclosed shopping malls, industrialized textile factories, even the motion picture industry would not exist without A/C.

To dive deep into this topic, check out this fascinating, obscure book: google.com/books/edition/… Essentially, HVAC enabled the motion picture industry (theatres), textile factories in humid environments, air travel, modern office environments, enclosed shopping malls…a much more impactful innovation than the Internet.

Interesting paper. This reminds me of Ritz-Carlton’s old MR. BIV initiative (Mistakes, Rework, Breakdowns, Inefficiencies, and Variation). We operated from the belief that guest loyalty was downstream of organizational reliability, and organizational reliability was largely a function of how effectively we reduced entropy in the system. What’s interesting about Chronos is not the conclusion that reliability matters, but its attempt to quantify the long-term economic impact of reliability improvements without waiting years of guest stays to observe the outcomes.

YES!!! This is a chapter for your next book. To expand on a few of your points: - consultancies will invest in partner level hires to place asymmetric options for growth. This new practice lead for “X” is a hope that their decades of experience will be the “5 mins” in the room to keep things in play (prospect engagement, internal resource alignment, etc). Most of their initial weeks and months of “work” will appear non-value add but it’s really about applying their mental model to a new organizational game board - there are MANY non value-added contributors inside the enterprise which drives up Coordination Tax. As a leader, you need to constantly monitor your meetings and comms to understand which of this corporate theatre is cosplay versus helping drive alignment, decisions, etc - young competent leaders can/will get promoted quickly based on their ability to navigate the internal dynamics and apply their own “5 minutes” of genius which reflects their mental model of how the org works

To clarify, I was looking at this from a different angle…not from a “let’s build a Large-E enterprise context graph” lens. Rather….if one is building a NewCo SW platform that is focused on “context graphing” an enterprise function (say, marketing campaign orchestration) then having this E2E observability is key to getting the context capture nailed. To make the example more visceral, a campaign is defined in Adobe Workfront and executed across a Martech and Adtech stack with 10+ channels, multiple ad agencies, multi reporting platforms, etc. The Workfront record is a snapshot / byproduct of hundreds of Slack/Teams convos, Otter/Zoom transcripts, and other contextual alignment which is “observed” but usually not synthesized in any coherent manner. Having a NewCo that has observes, catalogs, tag/classify these behind the scenes interactions to curated metadata and then applies to the campaign record is the business oppty. Large consumer brands are drowning in N million variants of campaigns and treatments. Until these N million campaigns and variants are able to be “context graphed”, then applying agentic scaling will be limited. Campaigns is just one example - hundreds more exist in key workflows.

Interesting thought experiment which ties to the current chatter re: context graphs. This observability and file changes is the heart of @JayaGup10’s decision traces

John Suh @john_ssuh

2 weeks ago

Increasingly, I believe companies may need to be rebuilt from the ground up, where you have a single timeline of all observability + product metrics + file changes laid out in a retrievable system, like Datadog + Posthog + Google Drive + Slack (really unified filesystem of Claude

210 179 2K 632K 3K

“As a leader you don’t want AI to microwave your thinking….you want to slow cook things” Great tips for how to use AI in a more nuanced way for deck/presentation content creation.

Take whatever number of people you thought might be in jobs related to AI deployment in the enterprise and multiply it by 10. Then probably 10 again. A major topic that keeps coming up in talking to CIOs across enterprises of all sizes and industries is the implementation gap for getting agents to work at scale and organizations on mission critical work. As the task goes from implementing a chat system that’s basically an LLM plus search, to connecting to real production systems that both can deliver meaningfully better productivity gains but also introduces meaningfully more risk, a whole new set of work has to be done. You have to ensure the right level of protection of data, updates to access control controls, migration of legacy systems to common modern platforms, create observability across what agents are doing, implement new workflows, figure out the human in the loop moments, drive the change management of the new workflows, and more. Then, all of a sudden the model capabilities get updated and you have to do a set of the above steps over again. Half of what you’ve done is obsolete, and the other half needs to be upgraded to take advantage of new capabilities. Or, token budgets run hot and you have to peel off some of the workloads to lower cost models that will be more cost effective. But then you have to go through those same steps. Enterprise are trying to figure out what is the right set of roles to go and implement the systems in their organization to ensure that the workflows are actually being executed properly, ensure it’s not just slop being produced, and to make sure their organization remains safe and secure. Many companies are starting by repositioning existing IT talent in these functions, but there’s also a growing need for the equivalent of internal FDEs to go take on these tasks in an enterprise. The looks incrementally closer to software engineering than it does traditional IT implementation. Next, almost all AI vendors (labs and the software players) will have some form of next-gen FDE or Applied AI architecture functions to help support these use-cases. The benefit here will be these companies have an incentive to make their capabilities work well so they can bring best practices from a range of customers they’re seeing and directly from the product innovation. And finally, we’re seeing the rise of all new AI services firms or major parts of existing services firms move into AI implementation. Companies will often want to bring in ostensibly neutral players that can work across their tech stack but also have seen best practices across their vertical. There are going to be tons of new service providers that get launched to do this, and many will eventually go and disrupt (or get acquired) by the larger player. Either way, all told, we’re in for years of AI diffusion, and along with it tons of new roles and areas of work to be done to deploy AI at scale.

A reminder that transformative infrastructure rarely sells itself…from Barry Diller’s autobiography: “I was soon to learn that ABC was the best possible place for anyone to carve out a TV career in the mid-1960s. In those days, the three networks commanded more than 90 percent of all television viewing—but as the youngest of the three, ABC was the also-ran network. We were scrappy and adventurous, with a kind of nothing-to-lose sensibility that allowed us to take program risks that CBS and NBC would never do. CBS had been number one for many years and was referred to as the Tiffany Network. NBC, which had been founded on Toscanini concerts during the radio era, considered itself the quality network. It had been created by RCA primarily to sell radio consoles and then televisions, both of which it manufactured. By the time I arrived at ABC, NBC had just spent millions to adapt its programming to color; as a result, it was about to sell even more television sets.” Many parallels to today.

Guardrails, guardrails, guardrails…

Aaron Levie @levie

a month ago

This is true of all agents, not just coding agents. Probably the biggest challenge that most companies run into in their agent strategy is getting agents the right constrained context to work with for a task. Too much information or conflicting sources, and the agent can easily

64 42 377 72K 312

Glad to see we’re past the ninja and rockstar phase of job titles

🎭 @deepfates

a month ago

The new thing in San Francisco is no longer chief of staff or MTS. It's wizards. Everybody's got to have a wizard. If your company doesn't have a wizard and a 10,000 year cosmic plan you're ngmi. At some top startups each C-Suite exec has a wizard of their own

66 42 1K 168K 140

@davidsenra @tobi Norman doors is the classic example of afffordances discussed by DNN in the Psychology of Everything Things. Classic - once you read it and see it, you can’t unsee it.

But hey, this will all work out okay, right? /4

So why are they in this? Not because they've made peace with disruption. They're there for DEAL FLOW into their portfolios and client bases. The equity stake is the hedge against becoming irrelevant while appearing to embrace the thing that makes them irrelevant. OpenAI's goal is to shrink the consulting TAM which their partners built careers expanding it. /3

@OpenAI launched a Deployment Company today. $4B, FDEs embedded in enterprise, promise of "durable systems that don't need consultants coming back." The partner list is where the story gets complicated. 🧵

OpenAI @OpenAI

2 months ago

Today we’re launching the OpenAI Deployment Company to help businesses build and deploy AI. It's majority-owned and controlled by OpenAI. It brings together 19 leading investment firms, consultancies, and system integrators to help organizations deploy frontier AI to production

677 2K 11K 7.9M 6K

Excerpt from my "unpack this announcement" session with Claude: The Incentive Misalignment Problem DeployCo's stated goal is to build durable systems — AI infrastructure that runs reliably in day-to-day operations without ongoing human coordination. That's the product pitch: you get systems that work, not consultants who keep coming back. McKinsey's, Bain's, and Capgemini's core economics depend on clients not having durable self-sustaining systems. The engagement never fully ends. There's always a next phase, a next transformation, a next capability to build. If DeployCo actually delivers on "durable systems," it shrinks the consulting TAM over time. So why are they in the partnership? My read: they're there for deal flow into their portfolio companies and existing client bases, not because they've made peace with getting disrupted. They're buying a seat at the table to stay relevant, not to accelerate their own obsolescence. The financial return on the equity stake is the hedge. This means the consulting partnership within DeployCo is structurally fragile in a way that the press release doesn't acknowledge.

Both of you are hitting the enterprise symptoms at diff altitudes. @levie is describing the work. @mcuban, you are describing the entropy. Underlying is the root cause / structural gap... ...enterprises don’t actually have a system responsible for deciding. They have pockets of decision logic embedded inside tools, workflows, and teams. Agents don’t fix that. Hence, all of the recent excitement re: context graphs and decision traces. It’s not just standing up agents. It’s making decisions legible, making context computable, and making authority explicit across systems that were never designed for it. The model churn you're pointing out makes this worse. If decision logic is tightly coupled to models or tools, every upgrade turns into rework. That’s how you end up with model sprawl, inconsistent behavior, and rising cost of coordination. (Our modern ERP-upgrade saga!) We gotta separate decision logic from execution. For example, define: - what decisions exist - who or what has authority to make them - what context is required - what constraints apply Then let models, tools, and workflows plug into that. Without that layer, we’re going to get exactly what both of you are describing. A surge of implementation work followed by a lot of expensive unwinding.

Sadly, very few are thinking of orchestration in the manner you are speaking of. Most “agent orchestration” today is one of: - sequential chains - hub-and-spoke coordinators - brittle DAG execution - prompt routers - simplistic planner/executor loops Very little discussion exists around: - authority boundaries - arbitration - conflict resolution - evidence weighting - temporal consistency - degradation behavior - semantic interoperability - resource contention - policy enforcement - memory reconciliation Just spent two days with 10K Stripe conference attendees and most are just solving for narrow-banded silos.