AMD's Advancing AI 2026 Summit Highlights Hardware-Software Symbiosis and CUDA Alternatives

·1 views

AMD held its Advancing AI 2026 summit at the Moscone Center in San Francisco, offering free access to keynotes, tech talks, and vendor sessions. A day-one panel featured Chris Lattner, creator of the LLVM compiler infrastructure, who argued that hardware and software must be reasoned about together rather than in isolation. Lattner suggested that CUDA's dominance mirrors the legacy grip of GCC and can be bypassed through better architecture rather than direct competition, a philosophy behind his Mojo language at Modular. Panelist Ramin Hasani emphasized approaching AI problems at the algorithmic level before turning to hardware optimizations, and pointed to liquid foundation models as an area of interest beyond standard transformer architectures. The event reflected the broader industry trend of companies investing heavily in free developer education to drive ecosystem adoption.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Engineer explains how to stay in control when AI writes most of your code

A developer at Exequtech describes using Claude Code to build production software, including an offline-first Android app and migration scripts, while maintaining full engineering ownership. He warns that AI agents left unchecked tend to refactor uninstructed code, fabricate APIs, and produce output that is subtly wrong in ways that surface only weeks later. His approach involves precisely defining constraints before prompting, carefully reading every diff, and interrogating the agent's reasoning rather than passively accepting its output. He also acknowledges a personal trade-off: shipping more while coding less has caused some atrophy in core programming instincts, which he counters by hand-writing code when sprint schedules allow. He frames the key distinction as using AI as a tool under deliberate control versus becoming a passive passenger to its output.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AWS WAF and CloudFront KVS Enable App-Detected, Edge-Enforced Bot Blocking

A feedback loop architecture using AWS WAF Dynamic Label Interpolation and Amazon CloudFront KeyValueStore (KVS) allows applications to detect malicious behavior and delegate enforcement to the network edge. The application layer, which has full context such as repeated login failures, identifies threats and writes fingerprint data to KVS via a PutKey call. CloudFront Functions then read KVS on subsequent requests to challenge or block flagged fingerprints before they reach the origin server. To reduce false-positive impact, each KVS entry carries an expiry timestamp that automatically lifts restrictions after a set period, with a daily cleanup job removing stale entries. A staged approach — registering fingerprints as 'challenge' before escalating to 'block' based on pass rates — adds an additional safeguard against wrongly blocking legitimate users.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Microsoft and Databricks Enable Single-Copy Data Sharing via OneLake and Unity Catalog

Microsoft and Databricks have converged on a shared technical foundation — OneLake storage and Unity Catalog governance — allowing organisations to maintain a single copy of Delta tables accessible across both platforms. The integration, which reached broad availability by mid-2026, eliminates the long-standing need to duplicate data and maintain separate governance models for each platform. Three distinct mechanisms enable the interoperability: Mirrored Azure Databricks Catalog (generally available), OneLake External Locations for Unity Catalog (in beta), and the Publish to Fabric workflow. Together, these tools let engineering teams write data in Databricks while Fabric, Power BI, and other Microsoft tools query the same physical files without copying them. Choosing the wrong integration mechanism, however, can recreate the old two-copy architecture under a new guise, making deliberate design decisions critical.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How TDS Classic Controls Item Deployment, Updates, and Deletions in Sitecore

TDS Classic uses three deployment settings — Item Deployment, Child Item Synchronization (CIS), and Recursive Deploy Action (RDA) — to determine how items are handled in a target Sitecore database during a deploy. Item Deployment is a per-item property that controls whether an individual item is created, overwritten, or skipped, without affecting its siblings or descendants. CIS determines how deep TDS scans a parent item's descendants, acting as a gate for RDA, which governs what happens to Sitecore items not present in the project. These settings operate on different axes and never overlap, meaning misconfigurations can cause unexpected outcomes such as templates being overwritten or deletions not reaching the web database. The behavior described applies to TDS Classic 6.x on Sitecore 10.x, as well as late TDS 5.x releases.

0 comments Read more at DEV Community