Microsoft to unveil its own coding AI to boost Copilot
Reuters says Microsoft is preparing a homegrown coding model and other specialized AI for Build, as Asana buys StackAI and Groq lines up $650M for inference.
The research desk's take → Vision-language models mix up 'up' and 'far'; new benchmark exposes the bias30-second briefing
- Microsoft to unveil in-house coding model next week
- First-party models reduce partner reliance for Copilot
- Funding and M&A cluster around inference, orchestration
One-Line Summary
Big platforms are pulling AI capabilities in-house while enterprise workflow orchestration and alternative inference back ends accelerate.
Big Tech
Microsoft to unveil in-house coding model at Build
Microsoft is preparing to show an in-house AI coding model — software that suggests and writes code — at its Build developer conference in San Francisco, alongside new models for transcription, reasoning, speech and images, Reuters reports, citing The Information; Microsoft declines comment. 1
The push aims to increase usage of GitHub Copilot and lessen reliance on outside providers as Microsoft adjusts its ties with OpenAI; Microsoft shares rise nearly 3% on the report. 1
Copilot has mostly leaned on models from OpenAI, Anthropic, and Google, but competition is heating up as Anthropic’s Claude Code quickly rises to the top spot for AI-assisted coding, intensifying pressure to keep developers inside Copilot. 1
Reuters also notes Microsoft is exploring startup deals to diversify beyond OpenAI and has a stated goal of building a cutting-edge AI model by next year; watch whether these new models land directly inside Copilot and how developer satisfaction and adoption trend after Build. 1
Industry & Biz
Groq seeks $650M to expand inference “neocloud”
Groq, which builds its own AI chips and systems, is reportedly raising $650 million from existing investors to grow its inference-focused cloud, according to TechCrunch citing Axios. 2
That follows a “not-an-acquisition” agreement with Nvidia valued at a reported $20 billion that moved senior Groq leaders to Nvidia and licensed Groq’s hardware tech; Groq is currently led by interim CEO Adam Winter and interim CFO Matt Eng. 2
TechCrunch adds that backers Disruptive and Infinitium agreed to fill the round if other investors don’t take pro-rata, signaling confidence in near-term demand for inference capacity; for product teams, more back-end choice can impact latency and cost. 2
Asana buys StackAI to run cross-system agent workflows
Asana, a work management platform, acquires StackAI, a no-code tool for building and governing AI agents that execute complex workflows across enterprise systems like ERP, CRM and ITSM; StackAI integrates with apps including Salesforce, AWS, DocuSign and Oracle via bi-directional sync. 3
Asana says pairing StackAI with its AI Teammates and AI Studio brings cross-system execution into the Asana Work Graph; in a proof-of-concept, Asana reports automating an SEO spend process by pulling live data across five marketing systems and handing actions to AI Teammates. 3
StackAI co-founders Tony Rosinol and Bernard Aceituno join Asana, and StackAI continues as its own product and brand; Asana positions itself as an “operating system for human-agent teams” with governed, multi-person agent workflows. 3
What This Means for You
For non-developer leaders who use or are considering GitHub Copilot, Microsoft’s in-house model push points to deeper integration and potential pricing/feature changes inside Copilot; align early with engineering on the 2–3 improvements that would most affect delivery speed and code quality. 1
If your organization runs on Asana, the StackAI deal signals agent-powered workflows that can read and act across core systems under governance — useful for support intake, IT requests, compliance, and marketing approvals where cross-app handoffs slow teams. 3
Teams building or buying LLM-powered products may see more infrastructure choice as Groq raises capital to scale inference; more back ends can change latency, throughput, and cost even when the user experience stays the same. 2
The broader pattern is consolidation: platforms you already pay for are working to keep AI work inside their stacks, which can simplify security reviews and reduce tool sprawl, while raising the stakes on vendor evaluation and exit options. 1
Action Items
- Set a Copilot check-in with engineering: Ask your engineering lead which Copilot features matter most and plan a short bake-off with current tools after Microsoft’s Build updates.
- Book an Asana + StackAI demo: Pick one cross-system process (e.g., SEO spend or IT request routing) and scope a governed, human-in-the-loop prototype with your Asana rep.
- Request inference back-end options: Ask your AI vendor if they support Groq or similar back ends and collect a one-page latency/cost comparison for your top workload.
- Draft an AI agent governance checklist: Define data scopes, approval checkpoints, and human-in-the-loop policies to apply to any Copilot or Asana agent pilot.
- Plan a 30-minute post-Build review: Summarize what Microsoft announces for Copilot and decide one pilot to run in your next sprint.
Terms in This Article
- Inference
Inference is the execution phase where a trained model receives new inputs and computes predictions, classifications, or generated outputs. For LLMs, prompts are tokenized, processed through prefill and decode, and emitted as tokens while latency, throughput, and memory usage become core operating metrics.
- OpenAI
OpenAI is an AI platform and API provider that offers models such as GPT‑5.5 and hosted tools to developers, exposing a unified Responses API and an Agents SDK so applications can perform text and multimodal generation, function‑calling tool use, and state‑preserving reasoning workflows, supported by safety guidance and operational documentation.
- Anthropic
Anthropic is an AI company that provides the Claude family of large language models and a developer platform, distributing pinned-snapshot model IDs across multiple access surfaces (first‑party API, AWS Bedrock, Google Vertex AI, and more) while supporting production use through a Transparency Hub and a Responsible Scaling Policy that disclose knowledge cutoffs, safety evaluations, and deprecation timelines.
Two desks on this story
Microsoft to unveil its own coding AI to boost Copilot
Vision-language models mix up 'up' and 'far'; new benchmark exposes the bias
A new arXiv study finds vision-language models entangle vertical position with distance, and releases a synthetic benchmark to expose the shortcut. Separate papers and repos focus on practical efficiency: confidence-aware KV eviction, online skill distillation for web agents, and a 100M-image permissive dataset.
Comments (0)