Rosa Del Mar

Daily Brief

Issue 59 2026-02-28

Investment Operations And Firm Mechanics In An Agent Era

Issue 59 Edition 2026-02-28 8 min read
General
Sources: 1 • Confidence: Medium • Updated: 2026-03-02 19:46

Key takeaways

  • Insight Partners attempted to start a European presence by hiring a well-known leader but shut it down after six months because the firm was not mature enough and needed tighter focus.
  • The main disruptive wave in AI will be driven by autonomous agents rather than by adding AI features to existing products.
  • Agent systems will add an orchestration layer that routes workflow components to different models based on task requirements and cost.
  • The long-term value of Salesforce as a system of record can be monitored via the health of ecosystem companies built on top of it, because displacement of those layers would imply reduced underlying value.
  • Agent-focused sandboxes may require materially lower latency (for example ~80ms rather than ~400ms) because agents can spin up extremely large numbers of sandboxes rapidly.

Sections

Investment Operations And Firm Mechanics In An Agent Era

  • Insight Partners attempted to start a European presence by hiring a well-known leader but shut it down after six months because the firm was not mature enough and needed tighter focus.
  • In the 2000–2001 downturn, the dot-com bust ultimately dragged down broader software valuations, not just dot-com companies.
  • Murdock characterizes his venture approach as 'go big or go home' and cites Twitter in 2009 as a high-conviction investment executed in under 30 days when the company had roughly 30 people and no revenue.
  • Murdock views Insight Partners' key breakout as surviving the post-9/11 tech collapse while many VC firms became dysfunctional or effectively zombies.
  • A new VC advantage will come from having exceptional data to quantify market white space and from evaluating founders partly by how effectively they use autonomous agents.
  • High-quality investing decisions require both logic and intuition, and autonomous agents will not have true intuition in the near term, so humans will remain necessary decision-makers.

Agentic Shift As Primary Disruption Vector

  • The main disruptive wave in AI will be driven by autonomous agents rather than by adding AI features to existing products.
  • Founders Murdock spoke with believe Cursor’s current product is obsolete relative to autonomous-agent-based coding workflows.
  • Several AI-native startups known to Murdock adopted autonomous agents to write code within roughly the prior two months, often after only weeks of rollout.
  • Legacy products with bolt-on AI can still succeed tactically, but AI-native companies will be structurally stronger as the market shifts.
  • Enterprises may take a year or more to fully enable autonomous agents, but products not built for agent usage will face severe pressure within roughly 6–18 months as agents write software faster and cheaper than humans.
  • Billion-dollar single-person companies are plausible because autonomous agents can function as employees and expand what one person can execute.

Agent Stack Architecture And Model Commoditization Via Orchestration

  • Agent systems will add an orchestration layer that routes workflow components to different models based on task requirements and cost.
  • Autonomous agents will increasingly select tools and infrastructure by running probabilistic benchmarks across multiple options in parallel sandboxes rather than relying on developer preference.
  • A standardized agent stack analogous to the LAMP stack is likely to emerge from open source and accelerate agent-driven software creation.
  • As orchestration matures, open-source models will gain share and specialized ASIC chips with models placed on-chip will become more important due to cost and tunability advantages.

Enterprise Software: Systems-Of-Record Fragility And Ecosystem Monitoring

  • The long-term value of Salesforce as a system of record can be monitored via the health of ecosystem companies built on top of it, because displacement of those layers would imply reduced underlying value.
  • Systems of record can become far more valuable or effectively valueless depending on whether they successfully capture major transitions such as tokenization rather than being bypassed by new records.
  • Enterprises may take a year or more to fully enable autonomous agents, but products not built for agent usage will face severe pressure within roughly 6–18 months as agents write software faster and cheaper than humans.
  • Software purchasing and usage will increasingly be performed by autonomous agents that act like employees with credentials, identity, and performance/spend reviews.

Infrastructure Bottlenecks For Agent-Scale Execution (Sandboxes, Latency)

  • Agent-focused sandboxes may require materially lower latency (for example ~80ms rather than ~400ms) because agents can spin up extremely large numbers of sandboxes rapidly.
  • Autonomous agents will increasingly select tools and infrastructure by running probabilistic benchmarks across multiple options in parallel sandboxes rather than relying on developer preference.

Watchlist

  • The long-term value of Salesforce as a system of record can be monitored via the health of ecosystem companies built on top of it, because displacement of those layers would imply reduced underlying value.

Unknowns

  • What fraction of real enterprise workflows can current agent systems complete end-to-end without human review, and how quickly is that fraction changing?
  • How widespread is production coding via autonomous agents, and what are the measurable impacts on defect rates, cycle time, and total cost?
  • Will a de facto standardized open-source agent stack actually converge, and if it does, what components become the durable choke points (runtime, sandboxing, identity, tool APIs, evaluation)?
  • How common is multi-model routing in production agent deployments, and what is the observed cost/performance benefit relative to single-provider approaches?
  • Are the claimed sandbox latency requirements (and the specific example latency targets) necessary for agent-scale throughput, and what workloads drive them?

Investor overlay

Read-throughs

  • If autonomous agents become the main disruption, value may shift toward orchestration layers that route tasks across models and manage tool selection, reducing dependence on any single model provider.
  • If systems of record do not become the trusted control point for agent identity and actions, ecosystem layers built on top may weaken first, implying earlier signals of underlying platform fragility.
  • If agent scale execution requires materially lower sandbox latency, infrastructure built for rapid parallel sandboxing could become a gating factor for agent adoption and operating cost.

What would confirm

  • Production deployments show routine multi model routing with measurable cost or performance gains and automated evaluation or tool selection becoming standard operating practice.
  • Observable deterioration or consolidation among ecosystem companies built on a given system of record occurs ahead of changes in the core platform revenue trajectory.
  • Benchmarks or customer data indicate agent workloads create large numbers of sandboxes and that lower latency materially improves throughput, reliability, or total cost at scale.

What would kill

  • Enterprise agent systems rarely complete workflows end to end without human review and that fraction does not improve meaningfully over time.
  • Most production agent deployments remain single provider without routing because benefits are marginal or operational complexity outweighs gains.
  • Sandbox latency targets do not correlate with agent throughput or cost in real workloads, suggesting claimed performance bottlenecks are not binding.

Sources