Sarab Jamwal | Execution Systems

AI pilots succeed bcz decisions are reversible Production changes that When automation touches financial approvals or compliance, reversibility disappears The constraint is no longer accuracy It is decision ownership Automation scales execution Authority determines viability

@levie Governance and identity are showing up as expected. harder problem tends to appear once multiple agents operate in parallel across workflows. it’s less about access and more about decision ownership escalation paths, and how accountability is traced across interacting systems.

Had meetings and a dinner with 20+ enterprise AI and IT leaders today. Lots of interesting conversations around the state of AI in large enterprises, especially regulated businesses. Here are some of general trends: * Agents are clearly the big thing. Enterprises moving from talking about chatbots to agents, though we’re still very early. Coding is still the dominant agentic use-case being adopted thus far, with other categories of across knowledge work starting to emerge. Lots of agentic work moving from pilots and PoCs into production, and some enterprises had lots of active live use-cases. * Agentic use-cases span every part of a business, from back office operations to client facing experiences from sales to customer onboarding workflows. General feeling is that agentic workflows will hit every part of an organization, often with biggest focus on delivering better for customers, getting better insights and intelligence from data and documents, speeding up high ROI workflows with agents, and so on. Very limited discussion on pure cost cutting. * Data and AI governance still remain core challenges. Getting data and content into a spot that agents can securely and easily operate on remains a huge task for more organizations. Years of data management fragmentation that wasn’t a problem now is an issue for enterprises looking to adopt agents. And governing what agents can do with data in a workflow still a major topic. * Identity emerging as a big topic. Can the agent have access to everything you have? In a world of dozens of agents working on behalf, potentially too much data exposure and scope for the agents. How do we manage agents with partitioned level of access to your information? * Lots of emerging questions on how we will budget for tokens across use-cases and teams. Companies don’t want to constrain use-cases, but equally need to be mindful of ultimate token budgets. This is going to become a bigger part of OpEx over time, and probably won’t make sense to be considered an IT budget anymore. Likely needs to be factored into the rest of operating expenses. * Interoperability is key. Every enterprise is deploying multiple AI systems right now, and it’s unlikely that there’s going to be a single platform to rule them all. Customers are getting savvier on how to handle agent interoperability, and this will be one of the biggest drivers of an AI stack going forward. Lots more takeaways than just this, but needless to say the momentum is building but equally enterprises are acutely aware of the change management and work ahead. Lots of opportunity right now.

AI pilots usually fail for organizational reasons, not technical ones. In a pilot, model output stays advisory. In production, the same output may trigger financial approvals, compliance actions, or operational changes. At that point the question is no longer accuracy. It is authority. Who approves the decision. Who owns the exception. Who stops the system when it goes wrong. Automation increases execution speed. Organizations have to redesign decision ownership to keep up.

@levie Higher leverage usually expands demand, The constraint tends to move upstream. As engineering output increases, coordination load, review bandwidth, and decision ownership become the new bottlenecks. Capability scales quickly, Organizational absorption usually does not.

Here’s how this plays out. Software used to be too expensive and hard to write to automate most things. Now it’s vastly cheaper and faster to code. Thus, leverage has gone up dramatically, which means we’ll use software for far more. Leasing to more demand for engineering.

AI can increase individual productivity quickly. The harder transition is organizational. When fewer people operate larger automated systems, escalation paths and decision ownership become more critical than raw output. Automation expands capacity. Governance determines whether that capacity remains controllable.

Net hiring at most 1000+ person companies (AI labs are exception) is already zero or negative. Most CEOs know their companies are bloated and don’t need as many people, especially with AI driving a step change in productivity. Leaders will start by keeping headcount flat, but as AI capabilities compound and small teams outperform larger ones, major cuts are inevitable.

Blended product builder roles can reduce handoff friction. The challenge appears as organizations scale. When product, design, and engineering authority sit inside the same role, escalation boundaries and decision ownership across functions have to be very explicit. Removing coordination layers can increase speed. But authority design becomes more critical as teams grow.

PREDICTION: THE CPO ROLE, AS WE KNOW IT, WILL VANISH IN FIVE YEARS At young AI native companies, the traditional PM role is on the wane, replaced with a product builder archetype that’s a combination of Product, Design and Engineering. These companies will never hire a CPO. A separate product leader leads to too much cognitive dissonance when the IC roles doing the actual work are blending, extra overhead and imposes an unnecessary coordination tax on the product development organization. Five years from now, these companies will be the leaders and set the cultural tone for the next generation, so my prediction is that all tech companies will stop hiring for the CPO role in five years. There will be a singular product development leader at each org. Ironically, this new role might still be called the CPO, except they will run the entire product development org. CPTO is far too unwieldy of a title and only exists today to alleviate confusion. Career implication: early / mid career product leaders need to stop aspiring to become CPOs. instead, you need to develop a panoply of product development skills across all three disciplines (+ analytics), be able to fluidly navigate the roles, and become a product builder, period. Farewell, CPO! It was a good 15-20 year run for this role in tech. But like everything else, it’s time to evolve.

@49agents Review speed helps, But past a certain scale the constraint isn’t just human review bandwidth. It’s who has authority to approve, override, or halt the system when escalation happens.

@SJamwalExec this is the real constraint most automation posts miss. you can scale output infinitely but review capacity is human-limited. thats why i focus on tools that let me review faster rather than just execute faster. context windows and diff clarity matter more than raw speed

Automation scales output quickly. The constraint shifts to review capacity and decision ownership. When automated workflows begin touching financial approvals or compliance reporting, escalation paths must operate at the same speed as execution. Throughput can increase overnight. Authority models cannot. If escalation design lags automation, adoption stalls.

@danbenger @levie Agent autonomy will require clearer delegation and accountability models across enterprises.

@SJamwalExec @levie Hi Sarab, great insights on agent identity! At Tego AI, we're building solutions to help enterprises manage agent autonomy with security visibility and control. Would love to chat about how we can support these challenges!

Agent identities is going to be a super fun and hard problem for software in the coming years. Most agentic systems today assume that the agent can do everything the user can do, and just operate as an extension of that user. This has worked well and is how most auth has worked for cloud software, and it has made integrations super easy over time. But then comes along systems like Openclaw, and suddenly we get a new view into what becomes possible with agents that can operate on their own. And when you have many of them running in parallel. You start to work with them like a colleague, not just as an extension of what you do. But of course this now introduces an all new complexity than the traditional approach. What if you want an agent to access only a small subset of your data? What if you want an agent to have its own sandbox to operate in without any risk of a blast radius if it goes off the rails? What if you want to create an agent that can work with others, without you seeing everything it’s doing? For all of these cases, agents will start to need their own identities inside of platforms. To do this, we likely will need new mental models for how we delegate controls and access to them, how you handle authentication, who gets to manage them in an organization and so on. Lots to figure out in this space right now.

Bet sizing is a governance problem as much as a strategy problem. The hard part isn’t placing the bet. It’s defining upfront who owns the decision to double down or shut it down once incentives and sunk cost start pulling in different directions. Without explicit kill criteria, bets quietly become permanent line items.

Every company must make a few bets to grow beyond its core business. Sizing bets appropriately - and figuring out when to double down on a bet or kill it - is key.

@levie Capability may be close to unbounded. The constraint shifts to institutional capacity. When agents can decompose and execute complex workflows autonomously, the hard problem becomes review bandwidth, delegation limits, and escalation design across parallel execution.

It’s pretty clear that the emerging paradigm of agents will be like if you had a human expert in any domain, and they had all the capabilities of a top engineer who could use any tool (or the write their own on the fly) to complete any task, along with unlimited compute and a file system to work with. That combination of skills and technology primitives provides you with somewhat limitless capability in AI. You’re no longer limited by only what the model was trained on, or the inherent context window limitations. The agent will simply spin up subagents to work on component parts of the workflow, and get expertise as needed throughout the process. For all known types of tasks that are frequently repeated, they have quick access to existing skills and tools to complete their work. We’re already seeing this in a range of fields where skills are being written for agents to follow either domain-wide or company-specific processes. Doing legal analysis in a specific way, running financial models, processing spreadsheets for complex data work, generating PowerPoints, and so on. And for areas they’ve never seen before, they can simply write code on the fly to do the work one-off. Imagine pairing an industry expert with an engineer that can code up any custom script whenever it wants. Compute is your only limiter. This approach seems to cover a fairly wide range of knowledge work. Obviously the first space to benefit the most from this has been in coding itself, but it’s clear that this go across all other areas of work and even personal agents. Kind of wild.

Speed without boundary control creates volatility. Enterprises slow down not from caution, but from exposure to unowned risk. Velocity scales only when control is explicit.

AI pilots succeed in sandbox environments. Production rollout stalls when automated decisions touch revenue recognition or regulatory exposure. Model accuracy is acceptable. Authority boundaries are not. Adoption fails where decision rights are unclear.

High performers create local efficiency. Without aligned incentives, they increase global variance. Optimization at one node increases instability elsewhere. Talent amplifies system design.

Cash flow is the hardest signal to fake. The second-order question is institutional absorption. When new rails or AI-native companies scale rapidly, compliance, settlement, and cross-border governance layers have to scale with them. Revenue growth proves demand. Durability depends on how well the surrounding control infrastructure keeps up.

Stripe’s annual letter is out: stripe.com/annual-updates… The biggest highlight is relatively consistent every year: the Internet economy is growing faster than the rest of the economy. This has compounded for enough years that it is essentially _the_ growth engine in places.

Historically, productivity shocks increase output and demand. The friction usually shifts to coordination and governance capacity. As compute and automation expand supply, organizations have to absorb more decisions, integrations, and risk exposure. Productivity can scale quickly. Institutional capacity scales slower.

@TheGrowthLedger Exactly, Handoffs are visible. Waiting is not. What most teams miss is that each handoff also introduces a decision boundary. If ownership at that boundary isn’t explicit, latency compounds even when skill is high. Capacity rarely breaks first. Authority clarity does.

Enterprise friction rarely lives inside teams. It lives between them. Every additional dependency increases decision latency. Beyond a threshold, coordination overwhelms capacity.

@Lilly7862 Streamlining helps. But past a certain scale, friction isn’t about collaboration quality. It’s about authority clarity at the boundary. When ownership of cross-team decisions is explicit, latency drops structurally.

@SJamwalExec Streamlined collaboration is key when dependencies start to choke progress.

Productivity gains do increase demand. The friction tends to move upstream. As output expands, coordination load, review bandwidth, and decision rights become the new constraint. Organizations don’t just need more experts. They need clearer authority boundaries around the work those experts now scale. Capability expands supply. Governance determines whether the demand can be absorbed.

There are a reasonable number of roles in the economy that as the output of that profession can increase by an order of magnitude because of AI agents, there will be more more demand for the experts in that field vs. less. Coding is an easy example to process. The vast majority of companies in the world have no ability to hire software engineers to do work for them because a single software engineer previously was pretty limited in their impact. So the idea of building out a team for whatever the project was just was nearly infeasible. In a world where their output is 5X or 10X greater this is no longer the case. You’ll actually start to see demand for deep technical expertise across even more fields that were software adjacent. IT teams will need engineers because they’ll be automating more workflows in their company. Pharma will need more software engineers for complex data work. Physical products will go more and more digital and need more engineers. Small startups that didn’t have a shot of building complex software now will. There will be plenty of similar versions of this in other fields. If you’re a good video editor right now, the world is going to be demanding far more high end video content since you can produce high quality video far cheaper. Lawyers will probably increase because we’re going to come up with increasingly exotic terms in our contracts. And so on. There has never been a better time in history to be an expert in your field. The demand will only go up.