Cody Enterprise 3.0 vs FlashInfer 2.0

“The primitive here is honest and specific: a context retrieval layer that indexes across unlimited repos and pipes relevant code into whatever LLM you bring. That's a real problem — the moment your codebase spans more than one repo, GitHub Copilot and Cursor both go partially blind. The BYOM configuration is the right DX bet; it puts complexity in config where it belongs and lets the context engine be the actual product rather than a forced model subscription. The moment of truth is asking a question that spans three repos — if that actually works without hallucinating package boundaries, this earns its enterprise price tag. What I want to see is the indexing pipeline documented: how fresh is the context, what's the staleness model, and does it handle monorepos differently than polyrepos? Those aren't marketing questions, they're the whole product.”

“The primitive here is a CUDA kernel library for attention computation and KV-cache management — not a platform, not a wrapper, an actual low-level building block you can drop into vLLM or SGLang. The DX bet is correctness and composability over abstraction: they expose the knobs (speculative decoding thresholds, LoRA batching configs) without hiding them behind a config YAML that pretends the complexity doesn't exist. The moment of truth is swapping in the FlashInfer attention backend in an existing serving stack, and from what the repo shows, that's genuinely a few lines. The 40% latency claim needs a methodology cite — they show specific token generation benchmarks on H100s with prefill/decode separation, which is at least a real number attached to a real setup, not a vibe. This is infrastructure that a competent team could not replicate in a weekend; the CUDA work is deep and the speculative decoding integration is non-trivial. Ships because the craft is demonstrably in the kernels, not the landing page.”

“The direct competitors are GitHub Copilot Enterprise and Cursor with codebase indexing — and neither of them has Sourcegraph's decade of code search infrastructure underneath. That history is the actual moat, not the AI wrapper on top. Where this breaks: organizations with highly fragmented access controls across repos, where the context retrieval either over-fetches (security problem) or gets permission-gated into uselessness. The SOC 2 audit logs are table stakes for any enterprise deal in 2026, so calling that a feature is a bit rich — but shipping it before competitors formalized it matters. What kills this in 12 months: GitHub ships deeper Copilot Enterprise context natively and the org that was already paying for GitHub Enterprise doesn't want a second line item. Sourcegraph survives that only if the context quality gap stays wide enough to justify the cost.”

“Category is LLM inference optimization, direct competitors are FlashAttention-3, vLLM's built-in attention kernels, and NVIDIA's TensorRT-LLM — none of which are sleeping. The 40% latency claim is real in a narrow regime: it applies to specific decode-heavy workloads on Hopper-generation GPUs with prefill-decode disaggregation; swap in an A100 cluster doing long-context prefill and the number shrinks. What kills this in 12 months is not a competitor — it's NVIDIA shipping optimized kernels directly into cuDNN or the next-generation attention primitives landing in TensorRT-LLM, at which point the delta collapses. What earns the ship anyway: multi-LoRA batching at scale is a genuinely underserved problem that the big players haven't prioritized, and Together AI has production traffic to validate these claims against real workloads, not synthetic benchmarks. The open-source release is credible signal that they're playing for ecosystem, not just headlines.”

“The buyer is the VP of Engineering or CISO at a company with 200+ engineers across multiple repos — this is a clear, checkbook-holding persona, and SOC 2 audit logs are specifically the procurement unlock that moves deals out of legal limbo. That's a real wedge. The BYOM configuration is smart revenue-defensibility: Sourcegraph becomes the context layer that persists regardless of which model wins the next benchmark cycle, insulating them from the commodity model price war. The risk is the expand story — once they land an enterprise, what does deeper adoption look like? If it's just more seats, they're a seat-count business, and seat-count businesses get squeezed when headcount freezes. The specific decision that makes this viable is owning the index, not the model — the index is sticky, the model is not.”

“The buyer here is infrastructure engineers at companies running self-hosted LLM inference at scale — a real buyer with a real budget (GPU compute costs), not a vague enterprise persona. The open-source release is a distribution play, not a charity: Together AI captures value through their managed inference platform, where FlashInfer improvements directly reduce their per-token compute cost and become a credible differentiator in a market where Fireworks, Groq, and Anyscale compete on latency benchmarks. The moat question is the hard one — open-sourcing the kernel library means competitors can adopt it too, so the defensibility is execution velocity and production integration depth, not the code itself. What happens when NVIDIA ships this natively is the real stress test, and the honest answer is that Together AI's moat shifts entirely to their managed platform and the workflow integrations built on top of it. Still a ship because the business logic is coherent: they're using open source to build pipeline credibility while monetizing on the managed layer, which is a proven playbook.”

“The job-to-be-done is clean: get an accurate, context-aware answer about code that lives in more than one repository without switching tools or copy-pasting context manually. That's one job, no 'and.' Onboarding for enterprise is always an IT/procurement journey, not a 2-minute trial, so I won't penalize that — but the individual free tier needs to get a solo dev to a cross-repo answer in under 5 minutes or it never seeds the enterprise deals. The product opinion is strong: Sourcegraph has committed to the context layer being the product, which means they're not trying to win on model quality. That's the right call given their history. The gap is that 'unlimited repositories' as a marketing claim needs to be stress-tested publicly — if there's a practical ceiling at 50 repos or 10M LOC, that needs to be in the docs, not discovered during a pilot.”

“The thesis here is specific and falsifiable: inference compute will remain the dominant cost in LLM deployment for at least the next three years, and kernel-level optimization will continue to yield meaningful gains even as hardware scales. What has to go right is that the prefill-decode disaggregation architecture becomes the dominant serving pattern — if monolithic batching stays standard, FlashInfer's architectural assumptions become a liability rather than an asset. The second-order effect that matters most isn't latency reduction for Together AI's own platform — it's that cheap, reliable multi-LoRA serving changes the economics of fine-tuning. If you can serve 50 LoRA adapters off one base model at acceptable latency, the cost of domain-specific fine-tuning drops by an order of magnitude, which shifts power toward the fine-tuning layer and away from base model providers. FlashInfer is riding the prefill-decode disaggregation trend, and it's on-time rather than early — vLLM and SGLang have already moved this direction, which means the ecosystem is ready to absorb this rather than resist it.”