Snap Engineering
eng.snap.com · Company blog · on HN since 2020
Snap's engineering blog on Snapchat's infrastructure, covering microservices, service mesh, Android rewrite, QUIC, and ML platforms.
What HN saysHN links it for details of Snap's multicloud service mesh, which uses Envoy as a gateway, in threads about Snap's cloud costs and architecture.
18HN people shared it
4linked it in comments
18stories submitted
211story points
4 months agolast shared
2020people per year2026
Most-shared posts
- 1From Monolith to Multicloud Micro-Services: Inside Snap’s Service Mesh4 people · 24 points · first shared 2020 · discussion on HN
- 2Microservices at Snapchat3 people · 11 points · first shared 2021 · 3 comments on HN
- 3Rewriting the Snapchat Android App2 people · 5 points · first shared 2020 · discussion on HN
- 4QUIC at Snapchat1 people · 155 points · first shared 2021 · 45 comments on HN
- 5Code Search: How Agents Search Across Snap's Codebase1 people · 3 points · first shared 2026 · discussion on HN
- 6Introducing Bento, Snap's ML Platform1 people · 3 points · first shared 2025 · discussion on HN
- 7Training Large-Scale Recommendation Models with TPUs1 people · 2 points · first shared 2022 · discussion on HN
- 8Snap Developer Ecosystem1 people · 2 points · first shared 2021 · discussion on HN
- 9Privacy at Snap1 people · 1 points · first shared 2020 · discussion on HN
- 10Shipping Two Apps in One on Android1 people · 1 points · first shared 2020 · discussion on HN
- 11Migrating Snapchat's AB Pipelines to GPU-Accelerated Spark1 people · 1 points · first shared 2026 · discussion on HN
- 12Modernizing the PlayCanvas Back End Infrastructure1 people · 1 points · first shared 2022 · discussion on HN
- 13Snap Engineering1 people · 1 points · first shared 2020 · discussion on HN
- 14CodePal: Snap Built an AI Code Reviewer for the Age of AI-Written Code1 people · 1 points · first shared 2026 · discussion on HN
- 15Agent format1 people · first shared 2026
What HN people wrote when they linked it
freeqaz · 2025-10-20
These are the best additional bits of information that I can find to share with you if you're curious to read more about Snap and what they did. (They were spending $400m per year on GCP which was famously disclosed in their S-1 when they IPO'd)
0: https://chrpopov.medium.com/scaling-cloud-infrastructure-5c68a023e0f6
1: https://eng.snap.com/monolith-to-multicloud-microservices-snap-service…
mapme · 2022-10-21
I think this is an unhelpful comparison. Simply put, when users exchange messages on WhatsApp it causes minimal traffic to a WhatsApp data enter as it’s mostly p2p. No ads, no persistence.
Snapchat requires semi persistence for stories (viewable for 24hr) and millions of users can view a single story. When a celebrity posts a story and 5mil users view it that requires more engineering effort that p2p what’s app messages. Additionally, serving ads requires a large amount of compute (ad ranking/serving), storage (impressions), offline processing (Hadoop clusters are hard), engineering, salepeople, economists, and auxiliary systems to support advertisers buying ads, etc.
As an example, check out snaps blog post on ML and ads https://eng.snap.com/en-US/machine-learning-snap-ad-ranking
dargueta · 2021-06-25
Seems they use Envoy as the API Gateway https://eng.snap.com/monolith-to-multicloud-microservices-snap-service…
everydaynews · 2026-03-14
Interesting approach — treating agent definitions the way Kubernetes treats infrastructure. The core idea is that your agent's identity, tools, constraints, and execution strategy all live in a single .agf.yaml file, and any runtime (LangChain, ADK, whatever) can execute it via adapters.
The engineering blog goes deeper into the motivation: https://eng.snap.com/agent-format
A few things that stood out to me:
The schema is grounded in POMDP formalism, which gives it a principled foundation rather than being yet another config format
There's a "tighten-only invariant" for multi-agent systems — child agents can never exceed their parent's constraints. Governance gets stricter as you go deeper, never looser
It explicitly separates agent-level constraints from org-level policies. The runtime composes both at execution time
They position it alongside MCP (tools) and A2A (agent communication) — three complementary layers rather than competing standards
The JSON Schema is published, there's an interactive playground on the site, and the spec itself is vendor-neutral. Curious if anyone has tried plugging this into their own stack.