Salesforce Unlocks Agent Visibility, New Voice Agents Coming

Salesforce released Agentforce 3, adding a Command Center that lets executives monitor AI agent performance in real time. The upgrade is generally available now and ships with native Model Context Protocol support so agents can tap hundreds of business tools without custom code.

Agentforce adoption has jumped 233% in six months, with over 8,000 customers and results like Engine cutting case handling time by 15%. The release also hosts Anthropic’s Claude models inside Salesforce’s infrastructure, halves latency, and extends availability to Canada, the UK, India, Japan, and Brazil.

The new observability layer tackles the “day two” problems of tracking multiple agents and triggering human handoffs when needed. PepsiCo’s decision to roll the platform across its global operations shows AI agents are shifting from pilot projects to central enterprise workflows.

Synthflow AI raised a $20 million Series A led by Accel. The cash is earmarked for hiring, R&D, and opening the startup’s first U.S. office.

The Berlin company offers a no-code platform for HIPAA- and GDPR-compliant voice agents that plug into more than 200 enterprise tools. It already serves over 1,000 customers and has handled 45 million calls, processing about 5 million each month.

Synthflow grew revenues 15× last year and keeps more than 90% of its enterprise clients, showing unusual stickiness in a crowded market. Accel cited the product’s deep integrations and compliance focus as key advantages over heavily funded rivals like Sierra and Bland AI.

Scientists from the University of Geneva and University of Bern tested six prominent language models on widely used emotional-intelligence exams. The AIs chose the expert-approved answers 81% of the time, far above the human average of 56%.

ChatGPT-4, ChatGPT-o1, Gemini 1.5 Flash, Claude 3.5 Haiku, Copilot 365 and DeepSeek V3 all topped human scores across five EI assessments. When ChatGPT generated new test questions, human evaluators rated them comparable in difficulty and originality to the official items, showing a correlation of 0.46 with the originals.

Outside experts cautioned that multiple-choice vignettes favor pattern-matching machines and don’t prove genuine emotional understanding. They noted that while AIs like Brazil’s trucking assistant Aílton can detect stress with 80% accuracy, claiming machines are more empathetic than people stretches the study’s evidence.

Southeast Asia’s micro, small and medium enterprises are racing to embed generative AI into daily operations to stay alive against larger competitors. NUS Business School professor Jochen Wirtz says those that fail to adopt will be pushed out of the market.

Boston Consulting Group projects AI will add about $120 billion to the region’s GDP by 2027, and Lazada–Kantar research lists customer service as the top e-commerce use case. Indonesia’s Lita Global doubled monthly gaming events and increased weekly revenue 20% after integrating OpenAI models.

High usage fees keep adoption narrow, with even Lita Global paying roughly $2,000 a month for OpenAI API tokens. Gartner forecasts average genAI API prices will drop below 1% of current levels by 2027, lowering the cost barrier for smaller firms.

MIT Sea Grant is developing LOBSTgER, a generative AI system trained exclusively on photographer Keith Ellenbogen’s underwater images to create realistic, scientifically grounded depictions of Gulf of Maine sea life. Itcaptures field data and convert it into machine-generated visuals that spotlight a region warming faster than 99% of the world’s oceans.

Custom code builds proprietary diffusion models that preserve Ellenbogen’s style and block outside data biases. The AI can both invent new biologically accurate scenes and enhance real photos, recovering detail lost to turbid water and challenging shooting conditions.

By fusing field photography with AI, the team reframes visualization as a public engagement tool rather than a technical exercise. Their stated aim is to extend the model across the Gulf’s full species range and later to other ecosystems, underscoring a hybrid future where observation and algorithm co-create conservation stories.