AI Daily Digest — Nov 3, 2026

TL;DR

6 takeaways

Salesforce signs a definitive agreement to acquire Cresta, whose voice agents handle customer calls for brands such as Marriott, Comcast and Alaska Airlines; terms were not disclosed, and The Information reports a $2.7 billion price
Sources: Salesforce, The Information
Traversal, which builds AI agents for site reliability engineering, announces a $120 million Series B at a $1 billion valuation, co-led by Kleiner Perkins, Redpoint Ventures and Khosla Ventures
Sources: Traversal
Meituan releases LongCat-2.5, a natively multimodal model with 1.6 trillion total parameters and about 48 billion active per token, available through the LongCat API at limited-time prices of $0.30/$1.20 per million input/output tokens
Sources: Meituan
Short: Snowflake opens a public preview of Cortex Training for serverless fine-tuning and RL on open-weight models
Sources: Snowflake
Short: Flower Labs introduces Endeavor 1.1, adding image input and an EU inference option
Sources: Flower
Short: SpaceXSI describes Data Bot, a Grok Bot Team Bot its data scientists teach for company-wide analytics
Sources: xAI

Story 01 / 03

What happened: Salesforce announced on November 2 that it has signed a definitive agreement to acquire Cresta, which it calls the voice-first operating system for customer experience. Cresta's voice agents resolve complex customer calls end to end, and the same platform guides human representatives in real time and analyzes every interaction. Salesforce did not disclose terms; The Information reports the price at $2.7 billion.
Why it matters: Salesforce says voice is where customers bring their highest-stakes issues. Kishan Chetan, EVP and GM of Agentforce and Service Cloud, said Cresta's agents handle complex, high-volume calls for some of the world's largest brands.
Enterprise impact: Salesforce plans to pair Cresta with Service Cloud so each conversation links to customer history and outcomes such as resolution and retention. It says Cresta will keep supporting the contact center and telephony platforms its customers use today.
Startup impact: Voice agents for the contact center are now a target for the largest CRM vendor, which raises the bar for standalone players on integration with customer records.
What to focus on: Closing is expected in the first quarter of Salesforce's fiscal 2028, subject to regulatory approvals.
Sources: Salesforce(opens in a new tab)The Information(opens in a new tab)

Comment with your GitHub account. One thread per story.

Story 02 / 03

What happened: Traversal announced a $120 million Series B at a $1 billion valuation on November 3. Kleiner Perkins, Redpoint Ventures and Khosla Ventures co-led, with Sequoia and Battery Ventures participating, and Kleiner Perkins and Sequoia added to earlier investments. The company builds AI agents that investigate incidents and find root causes in enterprise software.
Why it matters: Traversal argues that coding agents have sped up writing code while making production harder to keep running. It cites a survey in which 52% of developers say they now spend more time debugging code once it reaches production.
Enterprise impact: The founders come from causal machine learning and reinforcement learning research at MIT, Berkeley, Columbia and Cornell, and treat incident investigation as a causal reasoning problem.
Startup impact: Traversal says the money goes toward what it calls self-driving production, closing the loop between writing code and keeping it running.
What to focus on: Whether AI SRE tools can show measurable cuts in outage time at large customers.
Source: Traversal(opens in a new tab)

Comment with your GitHub account. One thread per story.

Story 03 / 03

What happened: Meituan released LongCat-2.5, a natively multimodal model with 1.6 trillion total parameters and about 48 billion activated per token, building on LongCat-2.0. Its API changelog lists the model as available on November 2, following a preview on September 25.
Why it matters: Meituan says it trained on more long-horizon trajectories and large-scale long-horizon RL, and turned failure trajectories from real agents into reproducible, verifiable training tasks. Images are represented as discrete tokens so the backbone handles vision and text in one form.
Enterprise impact: The changelog says LongCat-2.5 works with agent harnesses including Claude Code, Codex and Hermes.
Startup impact: Limited-time API pricing is $0.30 per million uncached input tokens, $0.006 cached and $1.20 output, which makes it a low-cost option to test for agent workloads.
What to focus on: Independent evaluations, since the published benchmark results come from Meituan.
Sources: Meituan(opens in a new tab)LongCat API changelog(opens in a new tab)LongCat pricing(opens in a new tab)

Comment with your GitHub account. One thread per story.

Snowflake Cortex Training preview
Serverless SFT and RL with LoRA or full-parameter training on open-weight models, deployed to OpenAI-compatible endpoints; Snowflake claims up to 61% lower cost than competing serverless offerings.
Snowflake(opens in a new tab)
Flower Endeavor 1.1
Flower Labs' third model this year adds image input and inference on EU GPUs; it reports GPQA rising from 92.0 to 94.4 over Endeavor 1.0.
Flower(opens in a new tab)
Data Bot at SpaceXSI
A Grok Bot Team Bot the data science team teaches with shared skills and reviewed memory; SpaceXSI estimates it handled about 70 hours of data-team work last week.
xAI(opens in a new tab)
CrowdStrike to acquire Fiddler AI and Dreadnode
CrowdStrike signed agreements to buy both AI agent security startups.
CrowdStrike(opens in a new tab)

Agents move into the contact center and production

Salesforce is buying voice agents for customer calls, and Traversal raised $120 million for agents that debug live systems.