One-Line Summary

Money is rushing into the AI inference and agent layers while large enterprises harden cost controls as AI moves from pilot to production.

Big Tech

Deutsche Bank says AI compresses projects to months, watches costs

Deutsche Bank says AI is speeding up technology work so projects that once took two years now finish in three to six months, a sign that large enterprises are turning AI into day-to-day productivity. A senior executive outlined the gains at the bank’s Bank on Tech event in Bengaluru. ¹

The bank adds that backlogs once measured in months are now cleared in weeks and that it has about 9,000 technology employees in India, around 45% of its global tech workforce. The scale highlights how global firms lean on their Indian hubs for higher-value software and R&D as they modernize with AI. ¹

Still, Deutsche Bank is imposing discipline: engineers get token quotas for AI usage and must show value to unlock more, reflecting a shift to usage-based pricing from vendors. The bank also favors simpler models for routine tasks and is building tools to extract and analyze financial data and link external events to portfolio exposure. ¹

Industry & Biz

Baseten reportedly lines up $1.5B at $13B valuation

TechCrunch, citing the Wall Street Journal, reports that AI inference platform Baseten is close to finalizing a $1.5 billion round at a $13 billion valuation, just five months after announcing a $300 million Series E at a $5 billion valuation and nine months after a $150 million Series D. The rapid sequence underscores investor focus on the inference layer. ²

The round is described as “split-priced,” with some investors buying at $13 billion and others at $11 billion; the jump implies roughly a 160% valuation increase in under half a year. The deal is said to be co-led by Spark Capital, Sands Capital, Altimeter Capital, and Wellington Management. ²

Baseten provides the layer that runs models after a user submits a prompt, routing requests to the best-for-task option and leaning on capable, lower-cost open-source models to manage spend. TechCrunch frames the moment as an “inference gold rush,” with capital concentrating on serving production workloads reliably and cheaply. ²

For buyers, this signals more multi-model routing offerings designed to balance quality and cost at scale, potentially expanding options beyond a single premium model. It also suggests more competition on inference pricing and service-level guarantees as players differentiate on speed, reliability, and total cost. ²

YC Demo Day: agents, infra and security draw premium valuations

TechCrunch spoke with eight investors to identify 11 standout companies from Y Combinator’s Spring 2026 Demo Day, noting at least two startups fetching valuations of $175 million or more. The set spans defense tech, robotics, AI infrastructure, developer tools, and AI agents. ³

Examples include 9 Mothers (AI counter‑drone, $1.6M in sales with a single contract projected to expand to $35M and a reported pipeline aim of $1B), Ploy ($27M seed to automate websites and growth), Superset (run 100+ coding agents simultaneously, compatible with any CLI agent and IDE), and Tasklet (connects to apps like Slack, Outlook, and Google Drive to perform tasks with natural language). Investors also flagged security player Silmaril and non‑engineer builder Lightsprint. ³

Allbirds becomes Smartbird as AI investing frenzy shows

A Reuters video roundup notes that Allbirds officially changed its name to Smartbird, marking a pivot from shoemaker to AI infrastructure, while shares of SpaceX soared, underscoring how AI narratives and investor appetite ripple across sectors. ⁴

The snapshot captures a broader capital rotation: more companies rebrand or reposition around AI, and high-profile aerospace names feature prominently in market talk tied to the AI boom. For operators, the message is that AI is now a board-level identity choice as well as a technology strategy. ⁴

New Tools

Gradial raises $65M to automate marketing workflows with AI agents

Axios reports that Gradial raised $65 million in Series C funding at a $675 million valuation to build an operating system where AI agents execute marketing work across tools companies already use, such as Adobe, Salesforce, ServiceNow, and Databricks. Customers include AWS, Prudential, T‑Mobile, Vanguard, Kaiser Permanente, and U.S. Bank. ⁵

According to Axios, T‑Mobile cut execution time for campaigns by 80%–90% with 99% accuracy, and regulated industries value encoding compliance rules into workflows so agents consistently apply requirements. Insight Partners led the round, with total funding now over $120 million. ⁵

Gradial plans to expand its roughly 100‑person team across engineering, sales, and marketing, pointing to growing enterprise demand for cross‑tool orchestration rather than single‑feature add‑ons. ⁵

What This Means for You

Capital is concentrating at two layers you feel at work: inference platforms that promise cheaper, faster model outputs and agentic systems that move work across your existing stack. Expect more multi‑model routing pitches (Baseten) and cross‑tool “AI glue” offers (Gradial) that target speed and cost per task. ²

Productivity gains are real but budgets matter: Deutsche Bank’s token quotas and usage monitoring are a practical blueprint for controlling spend without stalling momentum. If your team experiments with AI today, adopt similar guardrails and capture value evidence before expanding access. ¹

Marketing is an early win zone for agents because it touches many tools and approvals. If you pilot an agentic platform, start with one measurable workflow (e.g., keeping product pages current across channels) and wire in compliance checks from day one. ⁵

Early‑stage signals from YC point to rising attention on agent security and developer velocity. If your team relies on internal agents, track offerings that harden against prompt injection and consider tools that let non‑engineers safely ship small changes under engineer review. ³

Action Items

1. Set AI usage guardrails: Define token or request budgets per team and a simple path to request more with proof of value.
2. Book an agentic marketing demo: Trial an agent platform on one contained workflow (e.g., updating help content across channels) and measure time saved.
3. Compare two cheaper models on one task: For a repetitive task (summaries, drafts), test alternatives to your default model and log quality, time, and estimated cost.
4. Harden agent workflows: Add a written prompt‑injection checklist and require engineer review before agents push production changes.