<p><strong>TLDR:</strong> Two Postgres sharding startups landed fresh funding this week, with PGDog raising $5.5M and Supabase announcing a Series F to accelerate Multigres. Agentic code review data from four independent 2026 datasets is sobering: AI adoption quadruples code volume but defect rates jumped from 9% to 54% and review times are up 441%. Apple expanded Private Cloud Compute to Google Cloud and NVIDIA GPUs while keeping its full software control model intact. On the infrastructure side, AWS added durable storage to ElastiCache for Valkey with two consistency modes worth understanding before you reach for MemoryDB.</p>
<hr>
<h2>Four 2026 datasets converge: AI code quadruples volume, defect rates jump from 9% to 54%</h2>
<p>Faros AI, CodeRabbit, GitClear, and GitHub all independently measured the same thing this year: AI adoption roughly quadruples code output while delivering only about 12% real productivity gain. Defect rates went from 9% to 54%, review times are up 441%, and zero-review merges climbed 31%. The core problem is that agent-written code has no human intent behind it — reviewers have to reconstruct reasoning that never existed, which is fundamentally slower. The practical response depends on blast radius: solo projects can lean on AI reviewers and test gates, but teams maintaining large systems need tiered, evidence-required review with a human owning every merge. Running two AI reviewers with different strengths beats one, and keeping PRs small matters more than ever. <a href="https://app.daily.dev/posts/0UE8fcSA1" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>PGDog raises $5.5M, Supabase Series F targets Multigres as Postgres sharding heats up</h2>
<p>PGDog, an open-source sharding pooler currently processing 2 million queries per second across customers, raised $5.5M. Supabase announced a Series F with part of the capital earmarked for Multigres, their Vitess-equivalent sharding layer for Postgres. PostgreSQL 19 is also in beta with a few numbers worth knowing: 2x insert performance with foreign key checks and logical replication of sequence values. If you’re running Postgres at scale and haven’t looked at the sharding landscape recently, the options are multiplying fast. <a href="https://app.daily.dev/posts/P3a6V2AcJ" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Apple expands Private Cloud Compute to Google Cloud with NVIDIA GPUs and Intel TDX</h2>
<p>Apple is running Apple Intelligence workloads on Google Cloud Platform using NVIDIA GPUs and Intel TDX CPUs, while keeping its full software control model: stateless computation, cryptographically Apple-approved software only, no privileged runtime access. New hardware roots of trust include NVIDIA Confidential Computing and Google’s Titan chip. A cryptographically verifiable hardware ledger handles supply chain risk. All binaries will be published for public inspection, and security researchers retain access to live PCC nodes via the Apple Security Bounty Program. The architecture is a useful reference for anyone designing confidential compute pipelines on third-party infrastructure. <a href="https://app.daily.dev/posts/RDXkJbyeR" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>AWS adds durable storage to ElastiCache for Valkey with two consistency modes</h2>
<p>ElastiCache for Valkey now supports persistent workloads beyond caching. Synchronous mode replicates across at least two AZs before acknowledging writes; asynchronous mode acknowledges first and risks up to 10 seconds of data loss. A DurabilityLag CloudWatch metric tracks replication lag, and writes are rejected if the buffer exceeds 10 seconds. Targeted use cases include AI agent memory, RAG knowledge bases, and payment tokenization. The feature requires Valkey 9.0+ and raises a real question about when to use this versus MemoryDB — the overlap is significant enough to warrant a deliberate choice. <a href="https://app.daily.dev/posts/w3OiHX74d" target="_blank" rel="noopener nofollow">Read more</a></p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Coding agent harness engineering accounts for 10-17 SWE-bench percentage points, not model choice:</strong> A detailed breakdown of coding agent architecture shows Claude Sonnet uses 33K tokens in Claude Code versus 188K in Cursor for the same task, and SWE-bench scores vary by 10-17 percentage points across harnesses using the same model — the harness is doing more work than most teams realize. <a href="https://app.daily.dev/posts/biJg3iFvV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NOT IN with a NULL subquery silently returns zero rows in PostgreSQL; PostgreSQL 19 patch finally enables anti-join conversion:</strong> A single NULL in a NOT IN subquery discards every row with no error or warning — a 25-year-old planner limitation that PostgreSQL 19 commit 383eb21ebff finally addresses when both sides are provably NOT NULL; prefer NOT EXISTS on all current versions. <a href="https://app.daily.dev/posts/yvuFmd5XP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Actions TanStack cache poisoning attack illustrates shared-cache trust boundary risk:</strong> A breakdown of seven GitHub Actions misconfigurations includes the TanStack cache poisoning incident as a concrete example of shared caches being exploitable across trust boundaries, alongside pull_request_target pwn requests and unpinned third-party actions — the open-source zizmor scanner can audit workflow files. <a href="https://app.daily.dev/posts/RGaoXg12r" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Slack migrated 700+ EMR pipeline jobs from SSH to REST, eliminating SSH as a communication mechanism:</strong> Slack replaced SSH-based coordination in over 700 Amazon EMR data pipeline jobs with a REST-based architecture, improving security posture and operational reliability at scale. <a href="https://app.daily.dev/posts/31mThKudb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AVX-512 xor_gen() patch for Linux RAID achieves up to 43% performance improvement:</strong> Google engineer Eric Biggers posted a revised v2 AVX-512 implementation of the Linux kernel’s xor_gen() function used in RAID5/RAID6 parity and Btrfs, reaching up to 43% improvement over baseline — the patch is under review on the kernel mailing list. <a href="https://app.daily.dev/posts/hFXuR0l34" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>


Server-side development, system design, API patterns, and distributed architecture.

Backend Digest

A developer news roundup covering several major stories: four independent 2026 datasets show AI coding quadruples code volume but pushes defect rates from 9% to 54% with review times up 441%. Two Postgres sharding startups received fresh funding — PGDog raised $5.5M and Supabase announced a Series F targeting Multigres. Apple expanded Private Cloud Compute to Google Cloud using NVIDIA GPUs and Intel TDX while maintaining its stateless, cryptographically controlled software model. AWS added durable storage to ElastiCache for Valkey with synchronous and asynchronous consistency modes. Notable items include coding agent harness engineering accounting for 10-17 SWE-bench percentage points, a PostgreSQL 19 fix for the long-standing NOT IN with NULL subquery bug, GitHub Actions cache poisoning risks, Slack's migration of 700+ EMR jobs from SSH to REST, and an AVX-512 patch improving Linux RAID performance by up to 43%.

Postgres sharding gets $5.5M, agentic code review data shows 54% defect rates