Liveblocks AI Presence vs FlashInfer 2.0

“The primitive is clean: a React hook that injects an AI agent into Liveblocks' existing presence layer, giving it a cursor, a name, and a selection state — no new backend surface, no second SDK to wrangle. The DX bet is correct: they put the complexity in the abstraction, not in the integration. The moment of truth is a single `useAIPresence` call and your agent has a visible cursor within minutes. You could not replicate this on a weekend — Liveblocks' CRDT sync layer and multiplexed WebSocket infra are the actual hard part, and this just exposes a new participant type on top of it. The specific decision that earns the ship: they didn't add a new API, they extended the existing presence model — that's the right call architecturally.”

“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 competitor here is 'just log what your AI is doing in a sidebar,' which is what most teams ship today. AI Presence beats that because the presence metaphor maps to user mental models already trained by Figma and Google Docs — a cursor is legible in a way a log entry isn't. The scenario where this breaks is any app where the AI agent operates faster than human perception — a cursor flickering across a document at 200 tokens per second is noise, not signal, and Liveblocks hasn't shown throttling primitives in the demo. What kills this in 12 months: the underlying model providers build native multi-agent orchestration UIs and presence becomes a solved layer in the stack, not a differentiator. To be wrong about that, Liveblocks would need to own enough of the collaboration infra that switching costs make their presence layer the default regardless.”

“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 thesis here is falsifiable: by 2027, human-AI collaborative interfaces will require agents to express intent spatially, not just textually, because human coordination evolved around physical co-presence cues — gaze, gesture, position. If that's true, AI Presence is infrastructure, not a feature. The dependency is that AI agents remain slow enough relative to human attention that cursor metaphors remain meaningful; if agents complete work in under 500ms, the presence layer has nothing useful to show. The second-order effect nobody is talking about: this normalizes AI agents as social participants in software, not background workers, which shifts how users attribute responsibility and trust in collaborative outputs. Liveblocks is riding the multi-agent coordination trend and they are early — most teams haven't shipped a single agentic collaborator, let alone needed to display one. The future state where this is infrastructure: any SaaS with a collaborative canvas runs AI presence the way they run user avatars today.”

“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.”

“The job-to-be-done is singular and clear: make AI agent activity legible to human collaborators without building a custom observability layer. Onboarding survives the two-minute test if you're already on Liveblocks — the hook drops in and the agent appears; if you're not on Liveblocks, you're onboarding to an entire collaboration platform first, which is a different product decision. The completeness gap is real: this ships the presence primitive but not the interaction surface — users can see the AI cursor but the blog post doesn't address how users interrupt, redirect, or acknowledge agent actions, which means teams still have to build that layer themselves. The product has a clear opinion — agents are collaborators, not tools — and that opinion is the right one. Ship, but with the caveat that this is a primitive, not a complete human-AI collaboration solution, and teams should scope their expectations accordingly.”

“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.”