DEV Community: AI Tech Connect

Making Small Models Reliable at Tool-Calling

AI Tech Connect — Tue, 11 Aug 2026 12:01:15 +0000

Originally published on AI Tech Connect.

What actually breaks when a small model calls a tool Run a capable open-weight model in the 27 to 32 billion parameter range against a realistic tool suite and the transcript is rarely stupid. It reads the request, works out that a database lookup is needed rather than a calculation, and emits something that looks very much like a function call. The decision layer is fine. The failures cluster elsewhere: a date formatted as 11/08/2026 when the schema wanted 2026-08-11, a currency field carrying GBP_STERLING when the enum only accepts GBP, a nested object flattened into a string, a required field quietly omitted because the model could not infer it and did not want to say so. This is why "it picked the right tool" is the wrong success metric. Tool-selection accuracy is the easy half of the…

Read the full article on AI Tech Connect →

Verify the Model You Are Actually Being Served

AI Tech Connect — Tue, 11 Aug 2026 11:31:17 +0000

Originally published on AI Tech Connect.

What you need to know Here is the failure this guide exists to prevent. Your evals passed in March. You shipped, the numbers looked good, and you stopped watching. In June, support tickets start mentioning that the assistant has become oddly terse, or that it refuses a category of request it used to handle. Every call returned 200. The model field says exactly what it said in March. Nothing in your deployment history explains it, and by the time someone suggests the model changed underneath you, there is no way to prove it either way, because nobody measured the baseline. The root cause is a category error: treating the identifier a provider returns as though it were a measurement. It is a label attached by the party whose behaviour you are trying to verify — accurate about which alias…

Read the full article on AI Tech Connect →

Domain Expert to AI Engineer: A 12-Month Switch Plan

AI Tech Connect — Tue, 11 Aug 2026 11:01:20 +0000

Originally published on AI Tech Connect.

Why your domain is the scarce half Every applied AI system that survives contact with real users needs two ingredients: engineering that makes the thing run, and judgement about whether its output is correct. The market has built enormous machinery for the first and almost none for the second, because the second cannot be taught in a bootcamp. It is the pattern recognition of someone who has read ten thousand contracts, signed off two thousand discharge summaries or reconciled a decade of tax filings. The scarcity numbers are consistent. As of August 2026, one widely cited third-party aggregator estimate puts AI engineer demand up roughly 143% year on year, at a demand-to-supply ratio near 3.2:1 — about 1.6 million open roles against some 518,000 qualified candidates. Treat that as…

Read the full article on AI Tech Connect →

AMD Buys Taalas, the Chip That Hardwires One Model in Silicon

AI Tech Connect — Tue, 11 Aug 2026 10:31:19 +0000

Originally published on AI Tech Connect.

What you need to know AMD entered an agreement to acquire Taalas, a Toronto-based startup, announced in early August 2026 (CNBC, 6 August 2026). Taalas builds accelerators for inference, customised rather than general-purpose — each chip is hard-wired for a single AI model. Deal value was not disclosed in the reporting available. No price should be attached to this transaction. Almost nothing else is public — headcount, funding history, customers, chip specifications, process node and tape-out schedule were all undisclosed. The interesting question is the premise, not the price: committing a model into silicon only pays if that model stays relevant long enough to amortise the mask set and the design cycle. Small hardware acquisitions by large semiconductor firms are usually unremarkable,…

Read the full article on AI Tech Connect →

Sequoia's $1B Nuclear Bet on AI Data Centre Power

AI Tech Connect — Tue, 11 Aug 2026 10:01:17 +0000

Originally published on AI Tech Connect.

What you need to know $1 billion of equity, led by Sequoia Capital, with Sequoia partner Shaun Maguire joining the board — in a company founded roughly three years ago. A separate $200 million credit facility with Erebor Bank as administrative agent, alongside J.P. Morgan, Crescent Cove and Hercules Capital. Debt beside equity, at Series B. The milestone that preceded it: Ward 250 reached self-sustaining criticality on 18 June 2026 — the first time, Valar says, a company has taken a nuclear reactor critical outside a national lab. About a week later, Ward 250 powered an NVIDIA Blackwell GPU. What the money funds is manufacturing, not that reactor: the stated plan is to move from one working unit to producing fleets en masse through vertical integration, including on-site fuel production.…

Read the full article on AI Tech Connect →

Meta's Muse Glimmer Puts a 30B Agent Model on Your Laptop

AI Tech Connect — Tue, 11 Aug 2026 09:31:23 +0000

Originally published on AI Tech Connect.

What Meta actually shipped A 30B agent model under Apache 2.0 — released 10 August 2026, weights on Hugging Face at meta-models/Muse-Glimmer-30B. Apache 2.0 is an OSI-approved permissive licence, a different legal object from the Llama community licences Meta shipped previously. Built for agent work, not chat — multimodal input through a dedicated perception encoder, tool use and function calling, multi-step reasoning with failure recovery, 100-plus languages, variable reasoning-effort levels. Quantised to roughly 4-bit — the K-Quant-17GB variant keeps weights under 20 GB, with minimal to no degradation on agentic tasks by Meta's own account. DFlash speculative decoding — 3.1x on an RTX 5090, 1.8x on an M5 Max, 1.5x on an M4 Max. No published context window — for an agent model, the most…

Read the full article on AI Tech Connect →

Build a Deep Research Agent: Plan, Search, Verify, Cite

AI Tech Connect — Mon, 10 Aug 2026 15:31:16 +0000

Originally published on AI Tech Connect.

What you need to know The architecture has converged. GPT Researcher, LangChain's Open Deep Research and Milvus's DeepSearcher were built independently and all landed on the same shape: a planner, parallel search workers, and a publisher. The loop matters more than the search engine. A mediocre search used four times with reflection between rounds beats an excellent search used once. Verification is its own stage. Extract claims with their source attached, check claims against stored excerpts, and hunt for contradictions before a word of the report is written. Citations are a data-flow property. One number per unique URL, assigned by code, mapped claim-by-claim — not a formatting request buried in a prompt. Fully local is practical. An open-weight Qwen-class model on Ollama, local…

Read the full article on AI Tech Connect →

Roll Out Coding Agents Across an Engineering Team Without Chaos

AI Tech Connect — Mon, 10 Aug 2026 13:31:22 +0000

Originally published on AI Tech Connect.

What you need to know If you lead an engineering organisation in 2026, the coding-agent question has changed shape. It is no longer "should we let developers use these tools" — they already do, whatever your policy says. The question is whether the organisation converts near-universal individual usage into team-level throughput, or spends eighteen months generating more code, longer review queues and an unexplained cloud bill. The short version of this playbook: give the rollout a single owner, almost always the platform or developer-experience team, and a written AI-usage policy that is enforced by pipelines rather than goodwill. Start agents on work where failure is cheap — tests, small fixes, dependency updates — and graduate them to cross-module changes only when the data says…

Read the full article on AI Tech Connect →

Land Your First AI Conference Talk: A CFP Playbook That Works

AI Tech Connect — Mon, 10 Aug 2026 11:31:22 +0000

Originally published on AI Tech Connect.

What you need to know Most AI engineers who would make excellent speakers never submit a proposal, and most first proposals that fail do so for reasons that have nothing to do with the speaker's ability. The call-for-proposals process looks opaque from the outside, but it runs on a small number of mechanics that are entirely learnable. Four of them carry most of the weight. Selection is about fit, not worth. A programme committee is assembling a schedule for a specific audience, under constraints you cannot see. Rejection usually means "wrong fit for this room this year", not "not good enough". Your production work already contains the talk. The failure that cost you a weekend, the constraint nobody else has, the number you measured and nobody else published — that is the material. You do…

Read the full article on AI Tech Connect →

Indian AI Startup Funding Up 4x in H1 2026 — Still a Rounding Error?

AI Tech Connect — Mon, 10 Aug 2026 09:31:21 +0000

Originally published on AI Tech Connect.

What the numbers actually say $676 million across 57 deals — that is Inc42's tally for Indian AI-native startups in January–June 2026, up more than 4x from $162 million across 30 deals in H1 2025, with deal count up 90% to a six-month record. The average cheque nearly doubled — roughly $11.9 million per deal, against $5.4 million a year earlier. Bigger cheques, and far more of them. AI outran the wider market — the trackers cannot even agree on the direction of overall Indian startup funding in H1 2026: Tracxn logged $7.2 billion, up 12%; Entrackr $7.4 billion; Inc42's own count fell 9% to $5.2 billion. On all three, though, AI was the fastest-growing segment by a wide margin. The gap is still enormous — London's AI startups raised roughly $12 billion in the first seven months of the…

Read the full article on AI Tech Connect →

Gemini 3.6 Flash Hit GA in July: Google's Subagent-Tier Play

AI Tech Connect — Mon, 10 Aug 2026 07:31:27 +0000

Originally published on AI Tech Connect.

Three things builders should know The GA is real and dated. gemini-3.6-flash and gemini-3.5-flash-lite both reached stable status on 21 July 2026, per Google's own Gemini API changelog. If your team spent July heads-down shipping, this is the release to catch up on. Output got cheaper where it matters. 3.6 Flash lands at $1.50/M input and $7.50/M output — the same input price as 3.5 Flash but $1.50/M less on output. Google also says the model consumes 17% fewer output tokens than 3.5 Flash, which compounds the saving on verbose agentic loops. Flash-Lite is the actual subagent play. At $0.30/M input and $2.50/M output, Gemini 3.5 Flash-Lite is positioned explicitly for high-throughput, low-latency tasks such as agentic search and document processing — the jobs you fan out to workers, not…

Read the full article on AI Tech Connect →

Claude Code Turns Auto Mode On by Default: What Changes 14 August

AI Tech Connect — Mon, 10 Aug 2026 05:31:20 +0000

Originally published on AI Tech Connect.

What changes on 14 August Anthropic confirmed the switch in a blog post on 7 August: from 14 August 2026, auto mode is the default permission mode for new Claude Code sessions on Pro, Max and Team plans. Instead of pausing for approval before each tool call, Claude routes every action through a classifier and proceeds unless the action is judged "irreversible, destructive, or aimed outside your environment". When something is blocked, Claude looks for a safer alternative or asks you directly. Who is affected — Pro, Max and Team plans, new sessions only. Enterprise, the Claude API and sessions on AWS Bedrock, Google Cloud's Agent Platform and Microsoft Foundry stay opt-in for now; Anthropic says those will transition within about a month. Existing preferences survive — if you already set a…

Read the full article on AI Tech Connect →