What are the opening hours?

Monday 9am - 6pm Tuesday 9am - 6pm Wednesday 9am - 6pm Thursday 9am - 6pm Friday 9am - 6pm

The Infotech Report

Home
United States
San Diego, CA
The Infotech Report

Infotech.Report is an extensive resource library for the IT industry, bringing you the latest market studies, research reports and product information.

Get access to whitepapers, webcasts, case studies on an array of topics for the IT industry including data, enterprise applications, infrastructure, management, network and communications, security, servers, software and web development and storage.

08/22/2026

Salesforce has released Slack Code, a new service that brings AI coding agents into shared Slack channels for software tasks. The goal is to reduce handoffs and let stakeholders follow progress, add context, and intervene as work moves from idea to code.

When someone tags a supported agent, including Claude, Devin, GitHub Copilot, ChatGPT, or a Vercel agent, Slack Code creates a project-specific channel with relevant team members. The channel keeps conversation context, separates the agent chat, plan, and code diffs, and allows participants to review changes, give feedback, pause the agent, or stop it.

Once the task is finished, the agent can package the work for human sign-off and, in some cases, open a pull request. The channel then archives automatically as a searchable record.

Analysts say the approach could support faster fixes and better cross-functional input, while also raising concerns about visibility, notification volume, and security. How do you see shared AI coding workflows fitting into enterprise teams?

08/21/2026

Malicious Rust packages were published to crates.io on August 20, and the backdoor ran automatically during compilation. The compromised releases included [email protected], [email protected], and [email protected].

The packages introduced a dependency on proc-macro1, a typosquat of proc-macro2. That dependency used a build script to download and execute a second-stage payload during the normal build process.

Key details from the security findings include:
- arrayref has 245 million all-time downloads
- 53.7 million of those downloads occurred in the last 90 days
- the exposure windows were estimated at 86, 90, and 107 minutes

The payload could collect host, username, and operating-system information, inspect browser profiles, and establish persistence on Windows, macOS, and Linux. Organizations are being advised to check Cargo.lock files and local Cargo caches, then treat any affected build machine as compromised. How should teams respond to build-time supply-chain threats like this?

08/20/2026

The biggest failures in a fleet of scheduled AI agents did not come from model answers. They came from the layers underneath, including schedulers, shell assumptions, expired credentials, and safety checks that were never actually running.

One operator runs 49 scheduled AI agents on one laptop, with 24 more deliberately switched off. Over 18 months, model errors were expected and tested for. The unexpected outages came from issues such as a job never being registered with the scheduler, a command that worked in a terminal but not in a script, and a timeout guard that depended on a missing utility.

A quieter failure stood out even more. One agent could not access an email source and reported “unknown” instead of a false zero. That made the problem clear in seconds.

The lesson is direct. Read the running state, not the configuration file.

How often do systems look healthy while hiding a silent failure?

08/19/2026

DuckDB is preparing a major update, version 2.0, for release this fall. The next release adds client/server mode, asynchronous I/O, a new SQL parser, and a stable C API for extensions.

Key changes include:
- Client/server mode for single-client and multi-client use
- The quack extension moving to a stable release
- Faster query performance, with speedups of up to 40 times in some workloads
- Faster opening and reading of databases with large indexes and wide tables
- More detailed SQL error messages

The update also makes it easier for extensions to work across new releases without recompiling. Preview builds are already available in an unstable nightly version.

What feature matters most for database teams?

08/18/2026

GitHub restored services after a nearly 8-hour outage that disrupted Actions, pull requests, APIs, Git operations, Webhooks, Issues, and Copilot. The incident began at 1:40 PM UTC on August 17 and affected software development workflows across the platform.

At the height of the outage, GitHub reported about a 20% error rate across web experience and API traffic. Archive downloads and raw repository content downloads reached an error rate of approximately 50%, and SAML, OIDC, SCIM, and Team Sync were also affected.

Recovery was uneven, with some authentication problems lingering after other services returned. GitHub later said the incident was resolved at 9:15 PM UTC and that a detailed root cause analysis will be shared when available.

How should teams plan for outages that affect multiple connected developer services at once?

08/17/2026

Bonsai 27B is a 1-bit quantized AI model that shrinks a 54 GB model down to 3.9 GB, with open weights under the Apache 2.0 license. It is designed to support multi-step reasoning, structured tool calls, vision tasks, and computer-use agentic loops while keeping a 262K token context.

A local test on Windows 11 with an AMD Ryzen 5 3600 CPU, 32GB RAM, and an NVIDIA GeForce RTX 5060 showed clear trade-offs.

The model handled coding and reasoning tasks well, but it was slower than some other models. Output speed ranged from about 10 to 20 tokens per second on average, and disabling thinking improved speed but sometimes reduced accuracy or coherency.

Speculative decoding is supported, but the test setup did not have enough VRAM to use it.

Bonsai 27B shows strong promise for small-footprint AI work. How important are speed and model size in your own use cases?

08/15/2026

Nvidia has launched NeMo Switchyard, its new model routing option for AI workloads.

Model routing examines prompts and directs them to the most appropriate model, including the cheapest one able to respond effectively. The goal is to help requests run more efficiently while supporting more accurate results and lower runtime costs.

NeMo Switchyard provides a library for applying multiple routing approaches. It also gives developers a way to use a system-of-models approach and build more efficient, controllable agents.

Interest in model routing is growing across the AI sector. Cloudflare recently introduced a model router as part of a new suite of enterprise AI tools, and Stripe is looking to buy OpenRouter.

As AI spending keeps rising, routing options are drawing more attention. How do you see model routing affecting AI infrastructure decisions?

08/14/2026

Databricks is acquiring Electric to bring local Postgres databases closer to AI agents, with the goal of reducing latency in agentic applications.

Electric’s PGLite and Electric Sync will give developers a way to run a local Postgres-compatible database inside an application or agent environment while still synchronizing relevant data with a central database. Databricks says this two-tier approach can help agents work directly with data and keep shared, persistent data in Lakebase.

Analysts say the setup could make longer-running tasks faster and may reduce infrastructure costs by cutting remote database calls. They also point to new governance and security questions around local state, synchronization, auditing, and access control.

The move may give Databricks a temporary edge because none of its rivals currently offer the same WASM-Postgres capability. How significant could local state become in enterprise agent architectures?

08/13/2026

Many change management processes are heavyweight, but reversible changes can create room for a lighter review path.

A machine-verifiable change management process can reduce manual work by gathering evidence such as rollout and rollback scripts, audit records, uptime metrics, contract documents, scheduling conflict views, and notification records. The approach favors incremental automation instead of a full process rewrite.

The key idea is simple. If a change can be reversed, and partner systems can reverse or tolerate its effects, the team may be able to use a lighter review process. Destructive changes still require the full review pass.

The guidance also emphasizes presentation. Raw logs may contain the evidence, but review boards need legible summaries that are easy to reason about over time.

Which part of change management creates the most manual effort in your environment?

08/12/2026

Researchers said GitHub’s own event stream can act like behavioral telemetry for detecting software supply-chain attacks earlier.

At Black Hat USA 2026, Yossi Weizman of Microsoft and Mor Weinberger of Echo described an EDR-style approach built from GitHub webhooks, API data, and Git repository inspection rather than relying only on endpoint or network telemetry.

Their work focused on patterns seen across incidents such as Shai-Hulud, Trivy, and Megalodon. Those patterns included forged commit identities, poisoned tags, workflow abuse, OpenID Connect theft, and attempts to erase evidence.

They also introduced an open-source tool called GitHub Threat Detector, which reportedly includes 22 production detection rules and 12 beta rules. The tool is designed to correlate weaker signals into higher-confidence alerts.

What signals would be most useful for catching supply-chain abuse sooner?

Address

4660 La Jolla Village Drive Ste 100 & 200
San Diego, CA
92122

Opening Hours

Monday	9am - 6pm
Tuesday	9am - 6pm
Wednesday	9am - 6pm
Thursday	9am - 6pm
Friday	9am - 6pm

Website

http://infotech.report/

Alerts

Be the first to know and let us send you an email when The Infotech Report posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Business

Send a message to The Infotech Report: