SAP suspended most travel and hiring in July due to soaring AI costs, with exceptions only for AI-related activities and roles. The company cited token usage and related costs increasing as more AI-driven scenarios go live, and is rolling out a newly created AI tool company-wide. The freeze reflects a broader industry trend where companies are discovering AI deployment costs spiral quickly and are implementing usage restrictions and budget controls rather than realizing cost savings.
Todoist's parent company Doist adopted a philosophy of building AI features only when purposeful, rather than shipping every AI-enabled capability, with CTO Gonçalo Silva explaining that the company prototyped at least 18 AI ideas before settling on just a few to launch. The company plans to release Automations in August or early September, a feature that uses AI to interpret user requests into workflows but then executes them with traditional code rather than repeatedly invoking language models. By separating AI generation from reliable execution, Doist reduces costs, improves consistency, and maintains the ability to swap model providers without breaking user-facing behavior.
Amazon SageMaker Python SDK v3 now integrates generative AI inference recommendations directly into notebooks, allowing users to benchmark endpoints, generate deployment recommendations ranked by cost-performance tradeoff, and deploy optimized configurations without leaving their workflow. The new functionality in version 3.17.0 exposes operations like ModelBuilder.from_jumpstart_config(), start_benchmark(), generate_deployment_recommendations(), and deploy() to automate what previously required manual trial-and-error across instance types and framework settings. Users can now benchmark live endpoints, compare configurations like LMI vs vLLM, and iterate on deployment settings programmatically instead of manually testing multiple combinations.
Self-hosting AI inference makes financial sense above roughly two million tokens per day or when data sovereignty is required; below that threshold, hosted APIs are cheaper and require less engineering overhead. An MLOps engineer costs around 160,000 dollars annually, typically exceeding GPU hardware costs, making the salary the true expense of self-hosting. Most companies optimize through hybrid setups routing sensitive or high-volume work locally while using frontier models via API, achieving 40 to 70 percent savings versus all-API approaches.
DeepSeek-V4 Flash 0731 achieved 53.3% pass@1 on the DeepSWE coding benchmark versus GPT-5.6 Luna's 67.2%, but costs $0.10 per task compared to Luna's $0.61—a 6x price difference. When used in cascade (DeepSeek first, escalating to Luna on failure), the pairing solves 78.9% of tasks at $0.385 each, outperforming Luna alone on accuracy while costing 37% less. This strategy leverages DeepSeek's low cost to handle routine problems and reserves Luna's stronger reasoning for harder cases, fundamentally changing the economics of software engineering task automation.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
