Price:

BUSINESS

AI agents displace junior lawyers and engineers as firms reshape hierarchy

Tuesday, July 14, 2026 · from 3 podcasts
All-InAI Daily BriefDwarkesh Podcast
  • ElevenLabs hit $600M revenue without product managers, embedding engineers directly in HR and legal.
  • Ligora CEO says AI handles junior legal work, collapsing the $1 trillion billable-hour model.
  • Firms shift to fixed fees and success pricing as junior roles become AI managers.
  • Google's chaotic AI brand strategy is irrelevant against its 900 million users.

ElevenLabs CEO Mati Staniszewski runs a $600 million business without product managers. He embed engineers in non-technical departments like legal and talent. The engineers build custom automations and act as technical filters. This structure bets AI has made the middle-management coordination layer obsolete.

Staniszewski argues the bottleneck is no longer coding, but technical oversight. Engineers work in pods of five to ten people, moving directly from research to product. They ensure every internal tool has a built-in security check from day one. ElevenLabs deployed an inbound AI SDR agent to connect customers faster to the correct internal expert.

"The legal market is a $1 trillion behemoth, but only 4% of that spend currently goes to software. Ligora CEO Max warns that this ratio is 'bananas' and about to break."

- Max, All-In with Chamath, Jason, Sacks & Friedberg

The Ligora CEO notes elite firms like Kirkland & Ellis face a structural crisis. Their model relies on overcharging for junior associates to subsidize expert partners. AI can perform junior-level document review with 80% accuracy in seconds. The billable hour collapses. Savvy firms are shifting toward fixed fees or success-based pricing. The role transforms from researcher to manager of AI agents.

Ligora uses its own AI tools for in-house diligence. It closes a deal in 12 days from LOI. This contrasts with traditional lawyer incentives to extend timelines. The firm's data includes firm-specific precedent and a global repository of cases. It enables immediate 80% accurate responses for cross-jurisdiction queries.

"Gemini app surged to 900 million monthly active users this year. While specialized labs like Anthropic focus on the perfect developer environment, Google is simply saturating every digital interaction its users already have."

- Nathaniel Whittemore, The AI Daily Brief

Google's chaotic branding is a secondary concern. Distribution beats product design in the mass market. If the right AI tool is placed where a user already works, they won't care about the naming convention. Google is building a utility grid, not a single 'God model' interface.

A deeper tension is brewing between Google’s high-level research and its product teams. DeepMind CEO Demis Hassabis remains focused on a 5-to-10-year track toward AGI through world models. He is reportedly skeptical of the 'recursive self-improvement' path through coding agents that OpenAI and Anthropic sprint toward. Sergey Brin has formed a strike team to accelerate Google’s coding agent capabilities. This split identity explains why Google ships both Anti-gravity 2.0 for developers and Spark for consumers simultaneously.

The era of unlimited AI experimentation is ending. Token costs are now the primary concern for Fortune 500 CIOs. Google introduced 'compute-based usage limits' for its Ultra plans. The move signals a shift to a pay-as-you-go reality. The human voice is also becoming a licensable IP asset. ElevenLabs paid over $22 million to voice talent who license their clones on the platform.

The displacement is not limited to junior lawyers. AI coding agents are reshaping the entire engineering hierarchy. Firms that cling to the old coordination layers will subsidize inefficiency. Those that embed technical oversight directly into business functions will capture the velocity.

#AI Infrastructure#Regulation#Startups
Nathaniel WhittemoreDemis HassabisOpenAIAnthropicGeminiGoogle

Source Intelligence

- Deep dive into what was said in the episodes

All-In with Chamath, Jason, Sacks & Friedberg
All-In with Chamath, Jason, Sacks & Friedberg

The Trillion-Dollar Industries AI Is Disrupting: Voice, Law & the End of the Billable Hour • Jul 13

  • Mati reports that 11 Labs revenue growth accelerated sharply, reaching $100M ARR in 20 months, then $200M in 10 months, and $300M in 5 months.
  • Mati says 11 Labs now has 600 employees and maintains culture by embedding engineers in non-engineering teams like legal and talent for automation and security checks.
  • The company eliminated product managers, relying instead on cross-functional teams where AI elevates individuals from amateur to advanced level across coding, design, and customer understanding.
  • 11 Labs deployed an inbound AI SDR agent, finding customers provide more detailed information over a call, accelerating connection to the correct internal expert.
  • Mati observes consumers interact differently with AI voice agents, being more open about personal situations and quicker to cut off the agent without social guilt.
  • Mati outlines 11 Labs' safeguards against misuse: tracing all generated content, moderating voice and text inputs for scams, and providing tools to detect AI-generated audio.
  • Mati says 11 Labs' marketplace has paid over $22 million back to voice talent, enabling creators to license their synthesized voices, including for interactive content across languages.
  • 11 Labs partners with celebrities like Matthew McConaughey and works on restorative projects, such as recreating voices for individuals who lost them due to illness.
  • Mati positions 11 Labs as a platform agnostic to AI models, focusing on the interaction layer while competing with frontier labs on voice-specific models using specialized data and architecture.
  • JCal describes startups using ChatGPT for legal tasks like contract review and IP assignments, bypassing traditional corporate lawyers at early stages.
  • Joel argues AI is transforming law firm pricing models, moving from billable hours toward fixed fees for transactions or success fees in litigation.
  • Joel states Ligora uses its own AI tools for in-house diligence, enabling a deal to close in 12 days from LOI, contrasting with traditional lawyer incentives to extend timelines.
  • Joel says Ligora's data includes firm-specific precedent and a global repository of cases and legislation, enabling immediate 80% accurate responses for cross-jurisdiction queries.
  • Joel contends legacy legal research providers like LexisNexis and Westlaw struggle to pivot to AI-native models due to organizational politics, talent shortages, and slower operational tempo.
  • Joel asserts building general legal intelligence models is wasteful, but narrow models for specific tasks like tabular contract review can drive down cost and latency.
Also from this episode: (2)

Enterprise (2)

  • Joel explains the legal services market is a trillion-dollar industry dominated by manual service revenue, with only $40 billion spent on legal technology software.
  • Joel emphasizes compliance as Ligora's currency, noting the company hosts sensitive data for governments and weapons manufacturers without offering on-prem deployments.
#Enterprise#Labor#Startups#Regulation
The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis

Nathaniel Whittemore

How the Escalating AI Wars Benefit You • Jul 13

  • Whittemore argues Demis Hassabis's vision for AGI through world models and robotics diverges from OpenAI and Anthropic's focus on coding agents for recursive self-improvement, creating internal tension at Google.
  • Gemini Spark is described as a 24/7 personal agent built on Anti-Gravity, but its unclear positioning - citing both professional email drafting and small business customer service - confuses its target audience versus tools like Claude Code.
  • Anti-Gravity 2.0 rebrands Google's agentic coding harness as a standalone desktop app prioritizing the agent layer over the IDE, yet early reactions note its derivative feel compared to Codex and lack of surpassing Claude Coder.
Also from this episode: (6)

Big Tech (1)

  • Nathaniel Whittemore notes Google's AI strategy appears increasingly messy post-IO, yet its massive user ecosystem and Open AI's enterprise shift may grant Google a dominant position in consumer AI regardless.

Models (5)

  • Gemini Omni is positioned as a future anything-to-anything multimodal family, but its current release focuses on video generation with unprecedented editability, like changing scenes and character outfits, rather than raw quality.
  • Gemini 3.5 Flash benchmarks show competent but not state-of-the-art performance against Opus 4.7 and GPT-5.5, with its high token inefficiency making its speed focus questionable given the enterprise's primary cost concerns.
  • Google's product sprawl - including Omni, Spark, Anti-Gravity, Flow, Pix, and multiple Gemini tiers - creates user confusion, but its distribution via 900 million Gemini app users may render that confusion irrelevant for average consumers.
  • The Gemini Ultra plan price dropped from $250 to $200 monthly but introduced compute-based usage limits, reflecting a broader industry shift toward usage-based pricing as token costs dominate enterprise CIO discussions.
  • Whittemore recounts Google's AI history: the 2014 DeepMind acquisition created internal fragmentation, Bard's 2023 failure, Gemini's late 2023 consolidation under Hassabis, the 2024 AI Overviews debacle, and 2025's breakout with Notebook LM audio.
#Big Tech#Agents#Models#Enterprise
Dwarkesh Podcast
Dwarkesh Podcast

Adam Brown – A deep but accessible introduction to general relativity • Jul 10

  • Adam Brown argues the most important unanswered question in science is how the human brain achieves high sample efficiency and general capabilities with far less data than modern LLMs. His meta-level take is that neuroscience needs a technological power-up to answer it.
  • Brown's personal hunch is that AI has neglected complex, developmentally staged loss functions. Evolution encodes a specific learning curriculum through many different loss functions, which could be the key to the brain's efficiency.
  • Brown suggests the cortex might be an omnidirectional inference engine, predicting any subset of variables from any other subset, unlike LLMs which are natively optimized only for next-token prediction.
  • Brown outlines Steve Byrnes' theory that the brain's learning subsystem learns to predict the innate responses of a separate steering subsystem, wiring abstract concepts like 'spider' to primitive reflexes like flinching and enabling generalization.
  • Brown notes the human genome is only about 3 GB, a small fraction of which codes for the brain. This compactness is plausible if evolution mainly writes 'Python code' for specific reward functions and bootstrapping rules, not the entire learned model.
  • Brown says current LLM training uses a 'really dumb' form of reinforcement learning without value functions, which is surprising it works so well. In contrast, parts of the basal ganglia may implement simple model-free RL, while the cortex builds a model-based system.
  • A key disadvantage of biological brains is they cannot be copied or externally read, unlike digital models. Advantages include energy efficiency, collocation of memory and compute, and hardware co-designed for potential stochastic, sampling-based inference.
  • Brown states that creating a competent, misaligned agent like a 'paperclip maximizer' likely requires only minimal innate drives for curiosity and exploration, not the full suite of human social instincts. This is an alignment concern.
  • Brown advocates for massively scaling up neuroscience to get a 'ground truth,' specifically by driving down the cost of connectomics. The Welcome Trust estimated the first mouse brain connectome would cost billions; E11 Bio aims to reduce it to tens of millions.
  • Brown describes a moonshot idea of 'behavior cloning' or brain-regularized AI, where models are trained not just on labels but also to predict internal brain activity patterns. This could shape representations and improve generalization, but requires scalable brain scanning tech.
  • On automated theorem proving, Brown says RL from formal verification, as in Lean, will automate the mechanical parts of math. The harder challenge is automating the conceptual creativity of conjecturing interesting new theorems, which might require a loss function for explanatory power.
Also from this episode: (1)

Science (1)

  • Single-cell atlas data shows many more diverse and bespoke cell types in subcortical steering regions like the hypothalamus than in the cortex. Brown interprets this as evidence that evolution's genomic complexity is spent wiring innate reward functions, not the general learning algorithm.
#Models#Reasoning#Brain#Biology

Related Stories