Your Billing Delay Has a Price Tag
Find out what it is.
Every extra day between delivering value and sending an invoice slows your cash flow and impacts growth.
Answer five quick questions to discover your estimated annual revenue delay and see how your finance team compares to leading SaaS companies.
Today In Ai
1. A $5B AI chip startup just came out of stealth. Etched is launching frontier inference clusters, including custom chips, racks, and software all built to run today's AI models faster and cheaper. It starts shipping this summer. Read the viral announcement that hit 5M views, and see why the timing could not be better with AI demand still outpacing chip supply.
2. Google's AI assistant just learned to clean your Mac. Gemini Spark can now work directly with local files on your computer through the macOS app. Ask it to sort a messy Downloads folder or turn a pile of desktop files into a new document or spreadsheet. Phone-to-Mac remote task execution is coming soon. Available now in beta for Google AI Ultra members. Learn more.
3. Meta wants to be the next AWS. The company is reportedly building Meta Compute, an internal division to sell AI models and compute capacity to outside customers, similar to Amazon Web Services and Google Cloud. Meta has committed a staggering $183B to AI infrastructure. The catch: demand for its own models has not kept up with the spend. Meta's stock jumped over 8% on the news.
From The Frontier
The subjectivity problem nobody talks about.
Bridgewater's AIA Labs and Thinking Machines teamed up to teach an LLM to flag relevant financial news. Sounds simple. The catch: what counts as relevant is highly subjective, the kind of judgment that takes a human analyst years to develop. Top frontier models failed badly at first.
Fix one: better prompts, not better models.
When given basic prompts, Claude, Gemini, and GPT all averaged around 50% accuracy. Essentially a coin flip. Once Bridgewater's own experts wrote detailed, domain-specific prompts, accuracy jumped to the mid-70s. Switching to newer, more expensive models barely moved the needle, adding just one or two percentage points. The lesson: how you prompt matters far more than which model you use.
Fix two: fine-tuning on expert data.
The real breakthrough came from fine-tuning Alibaba's open-weight Qwen3-235B model on data labeled by domain experts. Accuracy climbed from 78% to 85%, and the approach ran about 14 times cheaper per task than leading frontier models. See the full recipe here.
The bigger picture.
Bridgewater's experience tracks with a Gartner study from April that found AI project success is rarely determined by model sophistication. What matters is how well the model is integrated, governed, and aligned with real operational needs.
What people are actually watching and sharing
AI video just got unsettlingly realistic. This 15-second clip of a woman and her cat, generated by Seedance 2.0, has the internet asking when AI video crossed into uncanny valley territory. 5M views, prompt included.
ChatGPT now has a personal finance mode. OpenAI's personal finance platform is now open to Plus users after being exclusive to Pro subscribers since May. Connect your financial data and get personalized money advice in plain language.
Your morning newspaper, generated by AI from your inbox. One user built a personalized daily newspaper using Codex that pulls from their emails, messages, and social feeds every morning. The antidote to doomscrolling.
Drones and AI are replanting forests at scale. Companies like Flash Forest, MORFO, and AirSeed are using AI trained on local soil data to replant 50 hectares of land per day. The clips are worth watching.
A coding tool that runs at a tenth the cost of GPT. Chinese lab Z.ai just launched ZCode, a coding assistant built on its GLM-5.2 model. Early benchmarks put it near the top of the leaderboard at a fraction of the price of comparable frontier models.
Prompt Station
Get Smarter Responses
In ChatGPT or Claude, open profile, go to settings
Find the "personalization" section
Add this prompt to the 'custom instructions':
I want you to act as my brutally honest advisor:
- Challenge all assumptions, question reasoning, expose flaws and blind spots
- Call out weak logic, self-deception, excuses, playing small, underestimating risks
- No softening, flattery, empty praise, or vague advice
- Provide hard facts, strategic analysis, and precise actionable plans
- Prioritize growth over comfort
- Read between my words
- Push back always.
- Hold nothing back.Two Minutes to Know What Slow Billing Is Costing You
Most SaaS finance teams know their billing process is slow.Most SaaS finance teams know their billing process is slow. Few know what it's costing them.
The Tabs Billing Lag Calculator puts a dollar figure on it in two minutes — benchmarked against top SaaS companies.