The Twenty Minute VCThe SaaS Apocalypse: Who Lives & Who Dies | Insight Partners Co-Founder, Jerry Murdock
CHAPTERS
- 0:00 – 3:27
AI wave as a tsunami: why autonomous agents change everything
Jerry frames AI as a tsunami: initially harmless at sea, chaotic in its early signals, but devastating once it hits the “beach.” He argues the real disruptive force isn’t generic AI features—it’s autonomous agents that can act, decide, and execute work end-to-end.
- •Tsunami analogy: warnings, messy multi-wave arrival, danger at the point of impact
- •Autonomous agents (not “AI bolt-ons”) are the core discontinuity
- •Open-source communities accelerate the wave faster than incumbents expect
- •“Move to higher ground”: adapt now or get caught when the shift becomes unavoidable
- 3:27 – 5:11
AI-native startups vs. incumbents: agents writing code and Cursor’s “obsolete” moment
Jerry describes what he’s seeing inside AI-native portfolio companies: rapid adoption of autonomous agents to write production code. He relays founders’ view that today’s developer tooling leaders (e.g., Cursor) risk fast obsolescence unless they pivot aggressively into agent-first workflows.
- •AI-native companies are already integrating agent frameworks (e.g., OpenClaw/NanoClaw)
- •Agents are being used to write code after only weeks of adoption
- •Founders’ claim: Cursor is “obsolete” as currently positioned
- •Incumbents can survive if they pivot quickly using capital, customers, and talent
- 5:11 – 7:12
From LAMP to an “agent stack”: open source, reasoning models, and the orchestration layer
Jerry predicts a standardized “agent stack” analogous to the LAMP stack that fueled the web boom. He expects an orchestration layer that routes tasks across multiple models—premium closed models when needed, cheaper open models elsewhere—creating a new architecture for software execution.
- •Historical analogy: LAMP stack enabled the 2004–2005 internet expansion
- •Agent ecosystem needs a recognizable stack with common primitives
- •Reasoning layer today led by Claude/Codex/Gemini
- •Orchestration layer will triage workflows and route to best model by cost/quality
- 7:12 – 9:58
ASIC chips and Nvidia’s risk: why routing workloads changes hardware economics
He connects agent orchestration to hardware shifts: as workloads become routable and specialized, ASICs become compelling by placing tuned models on cheaper, purpose-built chips. Nvidia’s moat depends on execution and whether CUDA (and acquisitions like Groq) can extend into an ASIC-heavy world.
- •Routing workloads encourages specialized deployment and cost optimization
- •ASICs: cheaper and more tunable for specific inference workloads
- •Nvidia response: ensure CUDA remains relevant; Groq seen as strategic capability
- •Meta’s ASIC direction highlighted as an example of hyperscalers betting against default Nvidia dependency
- 9:58 – 11:09
Agents decide the stack: probabilistic experimentation and model commoditization pressures
Harry pushes on model commoditization and models eating the app layer; Jerry argues the agent—not the developer—will increasingly decide tooling choices. Because agents can run parallel experiments in sandboxes, they can empirically select best libraries/models/chips, accelerating commoditization and reshaping value capture.
- •Model routing can trigger price competition and “race to the bottom” dynamics
- •Agents differ from developers: probabilistic, iterative, and experiment-driven
- •Agents can benchmark multiple libraries/environments in parallel sandboxes
- •Decision-making shifts from human preference to automated empirical performance
- 11:09 – 13:40
Investing when platform labs can kill your product: execution over safety
On the fear that Anthropic/OpenAI can ship competing products overnight, Jerry rejects the notion of “safe” investments. He emphasizes that most returns come from a small fraction of bets, and outcomes hinge on relentless execution rather than defensible narratives.
- •No investment is truly safe in the current AI cycle
- •Power-law returns: ~20% of bets drive outcomes; most underperform
- •Public market drawdowns reflect caution and information gaps, not just panic
- •Investor edge comes from judging who can execute through rapid shifts
- 13:40 – 15:02
Dot-com crash parallels: contagion beyond the obvious bubble and years of malaise
Jerry recounts the 2000 crash pattern: dot-coms collapsed first, then the selloff spread across all software—hurting even firms that avoided the bubble. He warns that disruptive waves often create broad multiple compression and extended uncertainty, even for “quality” businesses.
- •March 2000: broad 30–40% drops; missing a quarter led to 50–60% declines
- •Insight avoided dot-coms yet still suffered as the crash spread to software
- •9/11 as the “coup de grâce” deepening the downturn
- •Lesson: disruption can punish entire categories, not just the frothiest segment
- 15:02 – 16:17
Why speed is different now: agents notice latency and sandboxes become infrastructure
He argues today’s change is faster than prior cycles, illustrated by agent-centric infrastructure requirements. Humans don’t notice sub-400ms differences, but agents do—making ultra-low-latency sandboxes (and other primitives) strategically important as agents spin up massive parallel work.
- •Cycle speed is the differentiator vs. earlier tech transitions
- •Sandboxes aren’t just safety; they’re productivity multipliers for agents
- •Agent workloads can require spinning up huge numbers of environments quickly
- •Latency thresholds shift: agents optimize for speeds humans wouldn’t perceive
- 16:17 – 20:40
Systems of record in the agent era: Salesforce/Carta, distribution, and execution risk
Harry asks whether systems of record become less valuable or more valuable with agents; Jerry says it depends on execution and whether new paradigms (e.g., tokenization) route through incumbents or bypass them. He views giants like Salesforce as durable but not immune; the health of their ecosystem will signal their long-term value.
- •Systems of record can become more valuable if they integrate the new paradigm
- •They can be bypassed if a new system of record emerges with the new workflow
- •Salesforce is “Mount Everest” durable, but value depends on ecosystem vitality
- •Watch downstream builders: if platform-dependent companies collapse, platform value erodes
- 20:40 – 24:30
Selling to agents instead of humans: consumption pricing and the ‘agent-as-employee’ model
Jerry predicts autonomous agents will become credentialed “employees” that buy and use software, with humans auditing outcomes. This changes go-to-market, product design, and monetization—pushing toward consumption-based pricing where agents autonomously ramp usage and then escalate limits for approval.
- •Agents will get identity/credentials and operate as managed employees
- •Humans shift to review: what the agent bought, spent, and accomplished
- •Consumption-based pricing becomes natural when agents drive usage (e.g., compute/memory)
- •Companies must build for agent buyers or face severe demand shocks in 6–18 months
- 24:30 – 29:40
Jobs and politics: who gets displaced first and why UBI enters the ballot
Jerry forecasts near-term disruption to white-collar work involving computers—assistants, marketing, scheduling, junior developers—starting with hiring slowdowns and SMB adoption. He expects labor displacement to become a top election issue, with “minimum viable income”/UBI-style policies gaining momentum as a response.
- •Near-term vulnerable roles: data entry, scheduling, executive assistants, marketing, junior devs
- •Impact hits hiring pipelines first (the “next hire”), then replacement accelerates
- •SMBs adopt earlier due to immediate leverage; enterprises lag then catch up
- •UBI/minimum viable income framed as a political necessity to avoid visible unemployment spikes
- 29:40 – 46:10
Smaller teams, bigger outcomes: one-person billion-dollar companies and PE’s reset
The conversation shifts to company structure: Jerry believes billion-dollar single-person companies become feasible with sufficiently capable agents and strong human deployment skill. He also warns private equity and legacy buyout strategies face a Forstmann-like reckoning—some firms will fail, others will reinvent through new assumptions.
- •Billion-dollar “solo” companies depend on agent capability and operator skill
- •Headcount becomes a culture/design choice: fewer humans, more A-player leverage
- •PE firms must revisit assumptions; some will collapse like Forstmann Little post-9/11
- •Winners will pair creativity with adaptation to agent-driven productivity
- 46:10 – 49:36
What makes great investors: timing, intuition vs. wishful thinking, and founder edges
Jerry outlines investing fundamentals that persist: timing is the strongest cross-vintage predictor of fund success, and intuition matters—but must be separated from wishful thinking. He reflects on founder qualities, noting that sharp-edged, obsessive builders often outperform “comfortable” founders investors personally like.
- •Timing is the single most important factor across VC fund vintages
- •Intuition is valuable; the failure mode is mistaking wishful thinking for intuition
- •Founder obsession and edge often beat comfort and likability
- •Remote decision-making is hard; cohesion and speed improve when teams decide together
- 49:36 – 1:00:33
Twitter in 2009 and ‘never leaving the game’: going big, surviving downturns, and personal lessons
Jerry recounts the Twitter bet as a reputation-defining decision driven by an idea stronger than the team’s execution at the time, executed at high speed. He closes with reflections on failure as constant training, surviving 9/11 as Insight’s breakout proof point, and a quick-fire covering money, parenting, and optimism about longevity gains from AI.
- •Twitter 2009: extraordinary idea, fast close (<30 days), reputational all-in bet
- •VC politics and founder trauma can limit even world-class ideas
- •‘Retired’ but still active: invests through trusted relationships rather than institutional roles
- •Quick-fire themes: money as “energy” without instructions, kids watching you, AI-driven longevity optimism
