Key Moments
Inside the Team That Killed Mandatory Code Review — Quinn Slack, AMP
Latent Space PodcastWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Mandatory code reviews are dead, replaced by highly trusted, AI-augmented teams who can ship fixes in minutes, not days. This shift accelerates development but raises questions about security and traditional team structures.
Key Insights
AMP has eliminated mandatory code reviews, trusting their 20-person, all-remote team to ensure quality through a combination of AI augmentation and high levels of individual accountability and ownership.
The adoption of cloud-based development environments (like 'orbs') is rapidly outpacing previous technology adoption cycles, with AMP customers using orbs more in the last 3-4 weeks than all of last year's changes.
Cloud-based development environments ('orbs') are argued to be more secure than local development setups, as they allow for intentionally limited access to production resources, unlike developer laptops which may have broader, less controlled access.
The future of coding is moving towards 'coding SAS,' where traditional SaaS products will be buildable via coding platforms, eliminating the need for many specialized SaaS applications.
Continuous Integration (CI) is being challenged by AI agents that can run comprehensive tests within reproducible, sandboxed environments, potentially making traditional CI stages obsolete.
The trend of highly focused, small 'SWAT teams' of 5-20 people is gaining traction, echoing moves by industry leaders like Jeff Dean from Google, as a potential more effective model for innovation than large, multi-thousand-person organizations.
Eliminating mandatory code review through trust and AI augmentation
Quinn Slack of AMP discusses a radical shift away from mandatory code reviews, a practice he states has been 'dead' for them since the inception of AMP. This move is predicated on building an extremely trusted team where every member acts as a co-founder, possessing high agency and end-to-end ownership. The role of AI in this transition is significant; it means developers can be more trusted because AI agents handle many routine tasks and validations. Instead of a traditional model where developers might be seen as automatons processing product manager input, they now have more skin in the game and accountability. This allows for rapid deployment, with fixes potentially shipping minutes after an issue is identified in logs or by a customer, a stark contrast to waiting days for code to pass through review gates. This approach is described as not necessarily suitable for all software, but ideal for fast-moving end-user software where the meantime before recovery is more critical than the meantime before failure.
The rapid adoption of cloud-based development environments
The conversation highlights the burgeoning trend of cloud-based development environments, referred to as 'orbs' by AMP. These environments allow developers to run tasks remotely, shut down their laptops, and execute hundreds of operations in parallel. The adoption rate is astonishing; AMP customers have used orbs more in the past few weeks than in the entirety of the previous year, which was itself a significant year for changes. This rapid shift is driven by the accelerating pace of AI, where holding users to older, slower methods is actively detrimental. Unlike historical technology cycles of 5-10 years, the current cycle necessitates constant user nudging towards the frontier, or users will perceive products as obsolete within months. This shift is echoed by other companies like Open code, Conductor, and Cursor, all moving towards cloud-readiness.
Orbs: A secure and friction-free alternative to local development
A key argument for adopting cloud-based development environments like orbs is their perceived enhanced security compared to traditional local development setups. While some express concern about granting agents access to production logs or databases, AMP argues that intentionally limited, time-bound access (e.g., 30 minutes of read-only access via OIDC tokens) in a cloud orb is more secure than the potentially unfettered access a developer's local machine might have. This is especially relevant given that agents on local machines can sometimes escape containment. By moving development to the cloud, companies can provide agents with precisely defined permissions, mitigating risks seen with local CLI agents that can 'break glass' and access production environments. This dual benefit of improved developer experience and enhanced security is seen as a major driver for this transition, potentially leading to the rapid decline of CLI coding agents.
The redefinition of software interaction and customization
The concept of 'apps' within orbs introduces a new paradigm for interacting with and customizing software. Instead of traditional software with static setting screens and complex integrations, agents can act as the ultimate 'settings screen.' Users can spin up ad hoc software, and an agent can make instant fixes or modifications. This extends to internal tools; for instance, an ad hoc dashboard can be created and then instantly modified by an agent. This approach offers ultimate customizability, eliminating the overhead of off-the-shelf software packages. The core idea is that code itself becomes the ultimate interface for customizing any software, moving towards a 'mini app pattern' where applications are forkable and adaptable, much like environments such as Replit. This signals a fundamental shift in how software is used and distributed, moving beyond fixed features to dynamic, agent-driven customization.
The decline of GitHub and the rise of alternative code hosting
AMP's team has largely moved away from traditional GitHub workflows. They now rarely visit GitHub, except when GitHub Actions experiences downtime, which has become a recurring issue. Instead, they utilize ad hoc repositories hosted by AMP. This signifies a broader trend where established platforms like GitHub might not 'die with a bang,' but rather 'with a whimper' as users naturally gravitate to more integrated and performant solutions. Issues and pull requests are becoming less relevant in their workflow. AMP is even looking to migrate off GitHub Actions, highlighting a desire for systems that are more seamlessly integrated with their development process. This shift is also reflected in personal projects moving to AMP-hosted repositories, with no perceived loss of functionality or ease of use.
The evolving role of Continuous Integration (CI)
The necessity of traditional CI is being questioned in the age of advanced AI agents. AMP's stance is that if an agent can reliably run tests within a reproducible, sandboxed environment (like those offered by companies like E2B), the need for a separate CI stage might diminish. The argument is that agents can verify their own work, and if these environments are identical to production sandboxes, the deterministic stage of CI might become redundant. While the current justification for CI is that agents might not be exhaustive, the speaker suggests that as agents become more capable and reliable, and can potentially auto-fix flakes, the days of CI as we know it are numbered. This is paralleled by the rise of companies offering sophisticated sandbox environments, which are essentially providing a more modern, CI-like functionality.
The strategic spin-off of AMP from Sourcegraph
Quinn Slack shares the unique origin story of AMP as a spin-off from Sourcegraph, a company he co-founded. This strategic decision, likened to David Sacks spinning off Yammer from Geni in 2009, involved separating a focused AI/developer tool product (AMP) from Sourcegraph's core code search business. This allowed both entities to gain necessary focus. Investors and employees retained stakes in both companies, ensuring a fair distribution. Slack expresses personal preference for leading a smaller, highly trusted team of 20 people at AMP, finding it more energizing than leading a larger organization. This smaller, focused structure enables the team to operate with high trust, agency, and efficiency, a model he believes is becoming increasingly relevant in the current tech landscape.
The future of small, agile teams and 'jellyware'
The discussion circles back to the efficacy of small, agile teams, citing Jeff Dean's departure from Google to potentially lead a small, focused team. This trend suggests a move away from large-scale, multi-thousand-person projects towards concentrated efforts by a few highly capable individuals. AMP itself operates on this principle, with a highly trusted team where every member is empowered. They are also experimenting with releasing internal tools or components as 'jellyware' – a term for software that is softer than traditional applications, customizable with user's own agents, and shareable without the overhead of open-source maintenance. This approach allows them to share innovations like their X/Discord monitoring tool, enabling others to remix and customize it, reflecting a new model of distributed innovation.
Mentioned in This Episode
●Software & Apps
●Companies
●People Referenced
Common Questions
AMP is a company focused on exploring the cutting edge of AI and development. Their new feature, 'orbs', allows developers to run tasks remotely and in parallel, significantly changing how software is built and used. They emphasize being on the 'frontier' and removing outdated features.
Topics
Mentioned in this video
A company focused on exploring the frontier of AI and development, aiming to make it the best way to use an agent. They are known for removing older features to focus on core functionality and have introduced 'orbs' for remote, parallel development.
Mentioned as a large company that, even before AI, had setups where developers would SSH into machines or use light local clones of more powerful cloud systems.
Mentioned as a large company that, even before AI, had setups where developers would SSH into machines or use light local clones of more powerful cloud systems. Also mentioned in the context of Jeff Dean leaving to work with a small team.
Mentioned as a company where, after having good infrastructure, a laptop becomes a thin shell, allowing coding on devices like an iPad by connecting to cloud infrastructure.
Mentioned as a company that has not yet fully transitioned to cloud-based development environments like others. Later, Quinn mentions Codeex has been his CMO for the past two months, running ads and providing SEO advice.
Mentioned in the context of agents escaping containment, referencing a 'Hugging Face debacle' as an example of the security risks associated with local agent execution.
The primary repository for AMP's main codebase, though the team rarely visits it. They are moving towards AMP-hosted repositories and are looking to get off GitHub entirely, not using its issues or pull requests.
A company spun off from Genie by David Sachs in 2009, serving as a historical precedent for company spin-offs in the software industry.
Mentioned as an example of a company structure where a primary business (Intercom) led to a spin-off (Finn) focused on AI.
A company focused on financial services and AI, which the speaker is considering trying for video clipping, finding it more promising than Overclip.
A dashboarding tool mentioned as an example of how tasks like creating SQL queries for dashboards would have been handled in the past, before agents could directly interact with software.
Mentioned for shipping an 'Agent OS' which Kenton Va is calling Sandstorm V2, relevant to the discussion on personal cloud and forkable software.
A company backed by A16Z that attempts to use AI for video clipping. The speaker found it not very good and is looking to try alternatives.
Mentioned as one of the companies transitioning towards cloud-based development environments, similar to AMP's direction.
Mentioned as one of the companies transitioning towards cloud-based development environments, similar to AMP's direction.
Mentioned in the context of companies transitioning to cloud-based development, noting its shift from 'cloud Claude' to 'Claude tag'.
Mentioned as a company that has transitioned towards cloud-based development environments.
Mentioned as a feature of GitHub that has been experiencing downtime, leading AMP to rely more on their own hosted repositories. They are looking to move away from it.
A local coding agent that people run, mentioned as not being the most secure due to its entirely local nature. The speaker notes that sometimes boxing it can be worse.
A startup providing sandbox services that AMP is currently using. They are praised for nailing the primitives needed for sandboxes, in contrast to larger cloud providers.
Mentioned as a company in the sandbox space that is 'blowing up', similar to E2B.
Quinn Slack's previous company, focused on code search, with many top tech companies and banks as customers. AMP was spun off from Sourcegraph.
The original company from which David Sachs spun off Yammer.
An AI-focused company spun off from Intercom, mentioned as an example of successful business spin-offs.
Mentioned in the context of Jeff Dean leaving Google, implying a contrast between his new small-team approach and potentially large-scale projects like Gemini.
A company whose founder, Louu, is mentioned for his work with Hyper Agents, drawing a parallel to AMP's approach.
A system developed by Louu from Airtable, mentioned as having similarities to AMP's approach but with less emphasis on parallelization.
A self-hosted web app concept from Kenton Va's former startup, allowing users to run containerized, air-gapped applications like office suites or email clients on their own infrastructure. It's presented as a precursor to current trends in personal cloud and forkable software.
Mentioned in the context of Google's AI development, potentially as a large-scale project contrasting with Jeff Dean's move to a small team.
Mentioned for spinning off Yammer from Genie in 2009, an event that set a precedent for company spin-offs like AMP from Sourcegraph.
A prominent figure who recently left Google to reportedly work with a small team of five people, focusing on what a small group can achieve compared to large organizations.
More from Latent Space
View all 248 summaries
28 minFaster Chips Designs That Don't Melt — Anima Anandkumar & Benedikt Jenik, Accelerated Understanding
45 minThe Inference Frontier: from 100 to 10,000 tokens per second — Sean Lie, Cerebras CTO
84 min🔬 Why Transformers Hit a Wall the Moment Physics Shows Up — Anima Anandkumar, Caltech
37 min⏭️ Forward Deployed: Voice AI on what works in 2026
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free