Sep 25, 2025 · 1h 21m · latent-space
Amp: The Emperor Has No Clothes
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Quinn Slack and Thorsten Ball join the Latent Space podcast to discuss the inception, architecture, and high-velocity development of Amp, Sourcegraph's autonomous coding agent. They explore why maintaining extreme engineering agility, avoiding brittle scaffolding, and serving power users are essential for building frontier developer tools.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 24.5% of the talking time here. How this is scored →
The hosts as informed peer 4.5 Guest teaching 4.7 Guest disagreement 2.8 The hosts pushing back 2.1
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 40:45 Quinn dismisses prompt enhancement and bloated MCP tools
Quinn bluntly calls prompt enhancers bullshit features based on nuts theory, forcefully rejecting standard industry feature roadmaps.
Hardest push from the hosts ▶ 25:35 Alessio challenges revealing model names in product documentationAlessio presses the guests on why they expose specific underlying model details in their manual if models are supposed to be abstracted implementation details.
Biggest teaching moment ▶ 34:00 Thorsten on the illusion of deterministic LLM workflowsThorsten explains how non-deterministic models cause false confidence when wrapped in complex multi-agent architectures that fail unpredictably.
The host holds their own ▶ 49:31 Alessio shares empirical findings on model self-improvement toolsAlessio details his concrete experiment asking frontier models to build and utilize self-improvement tools, proving models frequently ignore extra tool harnesses.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| The Genesis of Amp and Breaking Free from Legacy Constraints | 4 | 4 | 2 | 2 | Alessio sets up the discussion around the transition from Cody to Amp. Thorsten and Quinn explain why creating an entirely separate product and unconstrained tool-calling architecture was necessary to avoid legacy enterprise baggage. | |
| High-Velocity Growth and the Fast-Moving Coding Agent Frontier | 3 | 5 | 3 | 1 | Quinn shares growth metrics and unit economics, while Thorsten frames the core thesis that the developer tooling landscape resets every few months. | |
| Engineering Velocity and High-Trust Organizational Structure | 4 | 5 | 2 | 2 | Alessio asks about team dynamics and balancing old enterprise contracts with a rapid startup pace. Thorsten describes pushing directly to main with no formal code reviews and leveraging existing security infrastructure. | |
| Product Architecture: VS Code Extension vs. Terminal CLI | 5 | 4 | 2 | 3 | Alessio asks about the architectural tradeoff of supporting both VS Code extensions and a terminal CLI. Thorsten explains the unexpected power of terminal composability and split panes over GUI extensions. | |
| Resisting Premature Optimization and Feature Creep | 5 | 5 | 4 | 3 | Quinn explains why Amp aggressively cuts secondary UI features and resists feature creep like custom buttons or BYOK pricing to stay focused on frontier performance. | |
| Scaffolding Models and Abstracting Model Selection | 5 | 5 | 3 | 4 | Alessio challenges the guests on why they even surface underlying model names in the manual. Thorsten and Quinn explain how harness scaffolding must go with the grain of specific model architectures. | |
| Foundation Model Dynamics and Multi-Model Routing | 6 | 4 | 3 | 3 | Alessio asks if proprietary model lock-in poses strategic risk. Quinn and Thorsten highlight their multi-model routing strategy leveraging Anthropic, OpenAI, and Google simultaneously. | |
| The Bitter Lesson of Agent Scaffolding and Non-Deterministic LLMs | 4 | 7 | 4 | 2 | Thorsten articulates the bitter lesson of agent engineering, cautioning that brittle deterministic workflows and complex MCP sub-agent graphs lead to massive developer hangovers. | |
| Vibe Coding Pitfalls, Token Inefficiencies, and Context Compaction | 4 | 6 | 7 | 2 | Quinn strongly criticizes prompt enhancers and bloated MCP tool sets as counterproductive bullshit features. Thorsten details the hidden loss of fidelity when using context compaction. | |
| Context Engineering and Specialized Sub-Agents for Testing | 7 | 4 | 2 | 3 | Alessio references context engineering research and shares his personal experience testing whether models actually use self-generated tools. Quinn details how sub-agents can handle specific test execution loops. | |
| Structuring Codebases and Logs for Agentic Consumption | 6 | 4 | 1 | 2 | The guests and host brainstorm optimizing software architecture and log outputs for agentic ingestion rather than human readability. | |
| Targeting Early Adopters Over Enterprise Laggards | 5 | 5 | 4 | 2 | Alessio discusses enterprise adoption inertia. Quinn argues that building products for laggards is a fatal distraction and that Amp strictly optimizes for frontier power users. | |
| The Outer Loop: Multi-Agent Parallelism and Git Review Friction | 6 | 3 | 2 | 2 | Alessio and the guests analyze the friction of parallel agent workflows and why existing GitHub code review paradigms are ill-fitted for agent-generated diffs. | |
| Unambiguous Specification and the Changing Shape of Engineering Teams | 4 | 6 | 5 | 2 | Quinn rejects the framing of non-technical users, arguing that anyone capable of unambiguous specification can command agents to produce immense enterprise value. | |
| Radical Trust and Founder Dynamics in Rapid Product Incubation | 3 | 4 | 1 | 1 | Alessio highlights Thorsten's career trajectory. Quinn describes empowering Thorsten with radical autonomy and trust to build Amp without standard corporate constraints. | |
| Evals vs. Vibes: Engineering Taste and Rapid Feedback Loops | 4 | 6 | 2 | 2 | Alessio queries their stance on evals versus intuition. Thorsten explains why formal eval suites are too slow for rapid product iteration compared to internal dogfooding and shipping twenty times a day. | |
| Looking Ahead: Building the Final Boss of Coding Agents | 2 | 2 | 0 | 0 | Brief wrap-up segment with hiring pitches and concluding thoughts on building the final boss of coding agents. |
Statements from this episode (34)
Slack: No AI developer tool has retained developers past 12 months
“There's not been any tool that has stuck with devs for more than, Six or 12 months or something.”
Slack: Amp is growing more than 50% month-over-month
“It's growing more than 50% month over month, a lot faster in, you know, some weeks.”
Slack: Two-to-three person teams reach six-figure run rates on Amp
“So we have teams with like two or three people that are on annual run rates of like hundreds of thousands of dollars.”
Slack: Asynchronous background agents will dominate AI inference and output
“That's going to be blown up with async agents when they're running 24, seven concurrently in the background, then you can have 10 or a hundred times as many, and that's going to dominate inference. That's going to dominate the output you get.”
Slack: Sourcegraph's Amp has positive gross margins
“It just so happens that we are positive gross margins along the way.”
Slack: Sourcegraph skips months-long compliance checklists for Amp using existing trust
“So there's a lot of things on AMP that we do, like no consistent pricing, no user model choice, no checking off all the boxes that security and compliance and legal want that, you know, takes nine months. We're able to get away without doing that stuff because…”
Ball: Amp core team has 8 people, ships 15 times daily without code reviews
“I think we're around eight people now on the AMP core team, and we still don't do formal code reviews. We still push to main. We still ship 15 times every day.”
Ball: Sourcegraph internal usage split 50/50 between CLI and VS Code
“We had, like, an internal poll last week at our company meetup where Beyang was asking who of you uses the CLI and who of you uses VS Code, and it was a fifty-fifty split”
Ball: Engineers under 25 find the terminal dated, preferring editor AI
“There's also a generational divide where I would say the younger people, you know, younger than 25, their terminal seems old to them and they were much more inclined to use the stuff in the editor.”
Slack: In-editor UI fix buttons will not matter in 6 to 12 months
“I don't think that that interaction is a really important way that people are gonna be interacting with AI in six months or 12 months.”
Slack: AI tools like Cursor and Claude Code will peak and decline
“A lot of these other tools that are great, like cloud code and codex and cursor and so on that they've forgotten what made them great and what made them grow so fast, which is building the very best product. And they built it in a way that's too overfit on the…”
Slack: Users of cheaper, mid-tier AI coding tools will churn within months
“There's absolutely a market today, literally today, someone will pay a monthly fee for that cheaper AI product today. But they're not going to be paying that in six months. It's going to be a different product, or they're going to be paying for something else.”
Ball: AI coding tools must optimize harness and scaffolding per model
“And people are now also waking up to the fact that it's not just the model, it's the system prompt, it's the tools, it's the harness. The scaffolding around the model. So I can give you the choice to use Gemini 2.5 in, in AMP, but without the system prompting …”
Ball: Foundation models will become background implementation details in AI tools
“So I think we're going towards a future where the model will become an implementation detail to some sense, and we will end up on a different abstraction layer.”
Slack: Half of internal Amp team uses a model other than Sonnet
“So we, in our internal team right now, and this could change. We have about half of our internal team using a different model other than Sonnet as their main way of using AMP.”
Fanelli: Cursor Default Model Switch Cost Anthropic $200M in ARR
“When cursors switch from Sonnet to GPT-V as like the default model that was like, you know, Two hundred million our revenue for Anthropic that kind of went away and like moved on to GPT-V.”
Slack: Top AI labs face a major customer stampede within two months
“I think we are one or two months away from a possible news cycle. That is the foundation model companies have spent billions of dollars in capex and hired like crazy. And now, you know, they're no longer the best in this realm and there's a huge stampede away …”
Ball: Agent scaffolding should be disposable as foundation models improve
“You want to build a scaffolding around the model, a wooden scaffolding that if the model gets better or you have to switch it out, the scaffolding falls away. You know, like the bitter lesson, like embrace that a lot of stuff might fall into the model as soon …”
Ball: Complex sub-agent workflows will result in user hangovers
“A lot of the features what we see, you know, where people build like elaborate workflows, like I have my custom slash commands and they trigger custom Custom sub-agents and they in turn trigger custom MCP tool calls behind which again another model is doing in…”
Ball: Coding agents should outsource typing, not engineering thinking
“The failure mode of outsourcing the thinking, but not the typing, which I think it should be the opposite. You still have to know engineering. You still have to know how to program. You still have to know your application and its architecture, how it's deploye…”
Slack: Competitors discounted products up to 100% to win deals against Amp
“So we've had one head to head loss with Amp where we lost against the usual players. And the reason why is one of them discounted their other product a hundred percent for two years. The other one discounted at 85% for two years, which is just crazy.”
Slack: AI prompt enhancers are a 'bullshit feature' that does not work
“Yeah, so prompt enhancer, that's a bullshit feature that doesn't actually work. The theory behind it is nuts, because what helps LLMs is not tricks and phrasing your prompt in a certain way. It's fundamentally information that you have in your head that you ca…”
Slack: MCP as user-facing tech creates high token costs and bad UX
“As a user facing technology, it is such a common failure mode where a user will go and add in some MCP servers. Auth is a huge pain, but let's say they get over that hurdle. Then they have, I don't know, 50 tools exposed that often are too low level granularit…”
Ball: Custom MCP tools fail if workflows diverge from frontier training
“If I give it this other custom-made MCP that we built internally, and our processes don't map to anything that OpenAI and Anthropic have seen or trained for, it won't be used, and you won't get good results.”
Slack: Amp only runs tests correctly 90-95% of the time
“Even with a good agents that MD guidance about how to run tests, I still see with AMP. And I think, you know, we've tried to make this really good. It only gets it right maybe 90, 95% of the time. Sometimes it'll run the wrong testing or it won't escape it cor…”
Slack: Most devs using coding agents lack proper Playwright automated testing
“The vast majority of devs who are building web applications with coding agents do not have play, right? And if they have it is set up in such a shitty way where it cannot really log into their app. They don't have any pattern for that.”
Ball: Unify and make logs verbose for AI agents, not human readers
“Instead of having the client log and having the browser log and having the database log, let's have one unified log, because then it's easier for the agent to just look at this log and make sense of it. And then it turns out it doesn't have to be nicely format…”
Ball: Web frameworks will integrate AI agents directly into the runtime
“What if the agent is integrated into the framework so that you can, if the application fails to run, you can ask the agent that has access to all of the context. And that's going to be more and more, I think like a lot of the, you know, developers will build s…”
Ball: Developers are actively modifying codebases to suit AI agents
“What we're seeing now with agents is as soon as somebody has seen what it can do, they have such a multiplying effect or this brings so much value that people are willing to adopt the code base for this. Like the first time in how many decades where people are…”
Slack: Users will tolerate bugs for cheap, personalized AI-generated apps
“People assume that coding agents need to meet the bar of writing the exact same kinds of software to the exact same standard, and that is not necessarily an assumption that end users, consumers will apply. If they have software that's much faster, cheaper, muc…”
Slack: Non-coding board member used Amp to replace $250,000 enterprise software
“And we've had non coders, including one of our board members who built something with AMP that replaced like 250 K a year piece of software that he used for a lot of their internal fund tracking.”
Slack: Vibe coding often results in poor feedback loops and quality control
“Vibe coding is this Really unproductive thing to discuss because everyone has a different definition of it. And too often it's having the agent write code with poor feedback loops and poor quality control.”
Ball: Engineers lacking business and product taste will lose value
“Like as an engineer, you cannot see yourself as the person anymore who types out a spec or turns a product PRD into code. I think you need to be aware of business. You need to be aware of a product. You need to know and have some taste for software. Otherwise,…”
Ball: The Sourcegraph Amp team does not use formal evals
“I think we don't have any set evals. We don't. And this was controversial up until a week ago, I think, when I think Boris from
Or two weeks ago from Anthropix that they don't have evals for the coding agent too.
But we don't, and we haven't had them.”