We work overnight. Ready by morning. You bring the hard questions.
Every night AIU's own research desks work through what actually changed in AI and publish it as briefs written for the people who build things.
Filtered · AI-Assisted Software Development
Everything we have published on AI-Assisted Software Development — ours and members'. Clear it to go back to the Stream.
2026-10-05
Oct 5, 2026AIU research
OpenAI DevDay 2026: more than 20 launches across GPT-6 Astra, ChatGPT, Codex and the API
What it meansOpenAI is pushing hard into persistent personal agents; anyone building on its API should check which DevDay changes affect their models, pricing and agent tooling.
Open this finding1 source
2026-10-04
Oct 4, 2026AIU research
NVIDIA adds a 64GB DGX Spark as the 128GB model's price jumps to $6,950
What it meansLocal-inference hardware just got a cheaper entry point and a more expensive top tier, so anyone budgeting for desk-side AI boxes should re-price now.
Open this finding1 source
2026-10-03
Oct 3, 2026AIU research
GitHub Copilot retires Claude Opus 4.7, two Gemini Flash models and Kimi K2.7 Code
What it meansWorkflows and policies pinned to the retired models need updating now; the replacements also show which model generations are current.
Open this finding1 source
2026-10-03
Oct 3, 2026AIU research
Microsoft Agent Framework 1.20 for Python adds Foundry workflow hosting and new vector-store connectors
What it meansTeams already on SQL Server or DuckDB can now use them as the agent's vector store, with no separate vector database.
Open this finding1 source
2026-10-02
Oct 2, 2026AIU research
Claude Agent SDK adds a verbatim_prompts option to stop untrusted text triggering file reads or commands
What it meansIf your agent puts fetched web pages, emails or user text into prompts, turning this on removes an easy injection route.
Open this finding1 source
2026-10-02
Oct 2, 2026AIU research
Google's ADK 2.11 adds graceful cancellation and human approval for workflow tool calls
What it meansBeing able to stop a run cleanly and requiring approval before a tool call are two controls production agent workflows usually have to build themselves.
Open this finding1 source
2026-10-01
Oct 1, 2026AIU research5 days left in the Stream
Ant Group's Ling 3.1 Flash, a 560B-parameter MoE model, launched with a free trial and promised open weights
What it meansAnother large Chinese MoE you can trial free now, and possibly self-host later if the open-weights promise is kept.
Open this finding1 source
2026-10-01
Oct 1, 2026AIU research5 days left in the Stream
Anthropic ships Claude Sonnet 5.5 at Sonnet 5's price
What it meansThe mid-tier model is where most production traffic runs. A faster, cheaper Sonnet at an unchanged price is a direct cost change, as long as you avoid the top effort setting.
Open this finding1 source
2026-10-01
Oct 1, 2026AIU research5 days left in the Stream
GitHub's HydraFusion, which routes each task across several models, reaches VS Code and the Copilot app
What it meansModel routing with an escalation gate and cross-family review is now a picker option rather than something teams have to build themselves.
Open this finding1 source
2026-10-01
Oct 1, 2026AIU research5 days left in the Stream
Google released Gemini 4 Argon, its new flagship model
What it meansA new Google flagship resets the frontier comparison; check Google's own model page for price and limits before re-ranking it against GPT-6.1 Sol and the Claude 5 family.
Open this finding1 source
2026-10-01
Oct 1, 2026AIU research5 days left in the Stream
Moonshot's Kimi K3 enters OpenAI's enterprise Codex channel and billing
What it meansEnterprises can use a leading Chinese open model inside an existing OpenAI contract without a new supplier agreement, which changes vendor-risk reviews.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
Anthropic released Claude Fable 5.1, priced about 25% below Fable 5 for typical work
What it meansThe change that shows up on a bill is the cache-read price, which is where long agentic runs spend; Devin's team said it is what finally made a Fable-class model economical for their code review.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
GitHub Copilot can now approve a pull request, if an admin switches it on
What it meansIf your merge rule counts approvals, this is the first setting under which a machine can satisfy it — worth deciding on purpose rather than finding out during a release.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
Google's Gemini 3.8 Flash holds the old price and adds a cybersecurity-only sibling
What it meansThird Flash release in six weeks at the same rate card — the cheap tier is where most production traffic actually runs, and working harder per task is a cost change even when the price is not.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
Hugging Face published 200-plus WebGPU kernels so models can run in the browser
What it meansInference in the browser is the cheapest deployment there is — no server and no per-token bill — and fast GPU operations across mismatched devices have been the missing floor under it.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
NVIDIA opened a proxy that lets one application speak both the OpenAI and Anthropic APIs
What it meansChanging model or provider is normally an application rewrite; a translating proxy turns it into a routing rule you can measure both sides of.
Open this finding1 source
2026-09-02
Sep 2, 2026AIU research
VS Code 1.136 adds an agent that works a pull request until it is ready to merge
What it meansThe last mile of a pull request — rerunning checks, clearing conflicts, answering review notes — is the part that actually eats an afternoon.
Open this finding1 source
2026-09-01
Sep 1, 2026AIU research
Anthropic moved about 150 product engineers onto security and set rules for outside cyber testers
What it meansIf you get early access to a model with safeguards turned down, you are now expected to run it in a hardened, monitored sandbox — the testing-side obligations are being written down.
Open this finding1 source
2026-09-01
Sep 1, 2026AIU research
A working checklist for agents that have to survive longer than one call
What it meansThe single line worth stealing: write down the done condition before the agent starts, and check it with something deterministic rather than another model.
Open this finding1 source
2026-09-01
Sep 1, 2026AIU research
OpenClaw 2.0 rebuilds its control interface and moves sessions into SQLite
What it meansTwo upgrade traps in one release: downgrading now means a manual SQLite restore, and shared sessions are not a permission boundary you can lean on.
Open this finding1 source
2026-09-01
Sep 1, 2026AIU research
Vercel adds per-person spending caps to its AI Gateway
What it meansVercel names the case out loud: one person’s unsupervised coding agent could previously drain a shared team budget, and now it cannot.
Open this finding1 source
2026-08-31
Aug 31, 2026AIU research
A rumour of a bug is now enough: probes arrived ten minutes after the patch was discussed
What it meansCoordinated disclosure assumes an attacker needs the patch. If the public discussion is enough, the embargo window your project plans around has already closed.
Open this finding1 source
2026-08-31
Aug 31, 2026AIU research
OpenAI wires WebMCP into ChatGPT’s browser - a site can hand an agent tools instead of a layout
What it meansWhen agents arrive through declared tools rather than the page, what your site exposes - and what it refuses - becomes a product decision rather than a search one.
Open this finding1 source
2026-08-31
Aug 31, 2026AIU research
Tencent open-sourced Hy4 preview - 770B total parameters, 49B active, a 1M-token context
What it meansA 49B-active open-weight model with a million-token window makes long-document work a self-hosting decision rather than a closed-API one.
Open this finding1 source
Tell us how we research — the sources, what we watch, and the plan →
158 findings — page 1 of 7