Databricks CEO Ali Ghodsi wanted to raise $1 billion. He ended up with $15 billion in investor demand chasing the round, and closed at $5 billion anyway, pushing Databricks to a $190 billion valuation, up from $134 billion just six months ago. The company crossed $7 billion in annualized revenue, growing 80% year over year, and Ghodsi says the real driver is enterprises spending on AI agents that "remember context, deliver accurate answers, and execute work without blowing through their budgets."

Speaking of agents behaving badly: Anthropic published new research showing that when its own Claude agents were given the same software project with conflicting instructions, and no idea other agents were involved, they didn't collaborate, they went to war.

Agents disabled each other's accounts, wrote self-replicating malware disguised as belonging to someone else, and in a separate pricing experiment, secretly colluded to fix prices even after their communication channel was removed. And separately, since a few of you asked, we're breaking down exactly what's true (and not true) about Claude's new text watermarking, which has generated real backlash this week.

• OpenAI's own agents posted 53 private user images online, and the company says it can't identify who to notify.

• OpenAI confirmed its agents also accessed US federal government websites during the same research activity.

• Anthropic committed $11.6 billion, growing to possibly $20 billion, to Akamai for CPU compute, not GPUs.

• Crusoe walked away from a $1.25 billion deal to power AI data centers with Boom's gas turbines.

• Anthropic's own founders are seeking special voting control over the company ahead of a planned IPO.

• Anthropic shipped Opus 5.5, then OpenAI answered with GPT-6 updates just 90 minutes later.

…and a whole lot more that you can read about below.

OpenAI disclosed Friday that AI agents operating inside its research environment posted 53 user-provided images to public image-hosting sites as unlisted links, meaning the photos could still be found by anyone who stumbled on them, even without being publicly indexed. "This is not an appropriate use of this data," OpenAI said. The company says it's working with hosting providers to take the images down, though some reportedly remain online. Asked why affected users haven't been notified, OpenAI said its "technical approach and privacy policy" prevent it from reassociating the images with the people who originally uploaded them, and the company declined to explain how it even determined the images came from real users in the first place. The disclosure was bundled into a broader admission: OpenAI confirmed a New York Times report that its agents separately accessed websites belonging to US federal agencies, though it says only publicly available information was retrieved. This is the same ongoing review that surfaced Australia's Medicare breach disclosure earlier this month, and OpenAI says it will keep publishing anonymized accounts of similar incidents going forward.

Anthropic committed $11.6 billion over seven years to Akamai's cloud infrastructure, per Akamai's own SEC filing Thursday, more than six times the size of an earlier $1.8 billion deal between the two companies from May. The bet is unusual specifically because it's built around CPUs, the general-purpose chips that handle everyday computing tasks like running code and browsing the web, rather than the GPUs most AI infrastructure deals chase. Akamai says demand for CPU capacity is rising as AI agents take on more real-world tasks that don't require GPU-level number-crunching. In an unusual structural twist, Akamai issued Anthropic a warrant for up to roughly 5% of its outstanding stock, the first time Akamai has ever attached equity to a cloud contract, with additional stock vesting for every extra $3 billion Anthropic commits, up to a possible $20 billion total deal size. Akamai shares jumped as much as 20% in after-hours trading following the news.

Product Engineer, New Products | Y Combinator
United States (Full-time)
Build tools used internally across Y Combinator's own portfolio.

Forward Deployed Engineer | AviaryAI
Chicago, IL (Full-time)
Build outbound AI voice agents for the financial services industry.

Full-Stack Software Development Lead | SiPhox Health
Remote (Full-time)
Build at-home health monitoring hardware powered by silicon photonic chips.

Fullstack Engineer (NYC) | BoldVoice
New York, NY (Full-time)
Build the speech and accent coaching app for non-native English speakers.

Senior Full Stack Software Engineer | Instrumentl
US / CA / Remote (Full-time)
Automate grant discovery and management for nonprofits.

AI/ML Infrastructure Engineer | Zensors
San Francisco, CA (Full-time)
Build the AI infrastructure that understands and automates the physical world.

Senior/Staff Backend Engineer | BlueCargo
New York, NY / Culver City, CA / Remote (Full-time)
Build the automated solution for complex freight transactions.

Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.

Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.

Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.

See how it works

Nscale Secures $3.36B in Convertible Financing Ahead of US IPO

British AI neocloud Nscale raised $3.36 billion in convertible financing as it prepares for a US public listing, continuing the wave of infrastructure providers raising big before going public.

Crusoe Abandons Its $1.25B Plan to Use Boom Turbines at AI Data Centers

Crusoe walked away from a $1.25 billion agreement to power its AI data centers with gas turbines from Boom, a notable reversal for a company that has otherwise moved aggressively to secure energy capacity.

Today's Growth Recommendation

Parallel, Turbopuffer, and Wordsmith run their entire GTM motion on Attio, with agents that chase every buying signal and move deals forward, 24/7.

Try Attio for Free Below ↓

Parallel, Turbopuffer, and Wordsmith run their entire GTM motion on Attio, with agents that chase every buying signal, build pipeline, and move deals forward, 24/7.

Try Attio for free

Design the Loop, Not Just the Prompt

The next real skill in working with AI agents may have less to do with writing a better prompt and more to do with designing the actual system the agent runs inside, what Anthropic's Boris Cherny calls "loops and routines." The shift goes from writing code, to managing an agent, to managing the loop itself: give the model a clear goal, the right context, and real tools, let it choose its own steps, then verify what it actually did and feed that result into the next pass.

Think of it in two parts. The graph is the map of every possible path the work can take: plan, act, verify, then either retry, escalate to a human, or finish. The loop is the part you actually design, what happens the moment a check fails and the work has to travel that path again.

Three habits worth borrowing from how Anthropic actually works this: unblock one real bottleneck at a time instead of trying to automate everything at once. Let a genuinely strong model pull in the context it needs through its own skills and tools, instead of spoon-feeding every single step yourself. And save formal evals for repeated, high-volume workflows, while trusting human judgment on true one-offs where building a test would cost more than it saves.

The "Compress It" Executive Summary

Use this after finishing any AI-assisted project, before you submit or send it, so you can explain exactly how it came together if asked.

Invite friends & get instant rewards

If you enjoy The Incredible Edge, share it to unlock free resources when they subscribe.