MongoDB has announced new capabilities for its Atlas platform that enhance context retrieval accuracy for AI applications and agents. The company introduced Automated Embeddings powered by Voyage AI, the Atlas Embedding and Reranking API, and voyage-code-4, a model designed specifically for code retrieval. Financial Times is using the technology to improve semantic search accuracy across more than 100,000 daily searches whilst managing costs. Legal AI platform Eve is employing the Atlas Embedding and Reranking API to surface relevant material during case work. The new features include automated embedding of documents as they're written, a unified API for embedding and reranking models, and vector search for streaming data. MongoDB's approach integrates retrieval capabilities directly into its operational database platform, eliminating the need for separate vector stores and synchronisation processes. The capabilities are now generally available.

Building your team's blog just got easier with Writizzy. Run your engineering blog with the whole team: roles, an optional review, inline comments and version history. Unlimited authors, on your own domain. Hugo Aug 3, 2026 · 3 min I have written a lot of blog posts for myself. Then I started a company, and inside that company I built the engineering team. I wanted Writizzy to adopt those same writing habits, because writing pays off in many ways. First, it is a great way to formalise things, and to explain them simply. As Boileau put it, "Whatever we conceive well we express clearly". Writing a good post forces you to dig into a subject and clarify your own thinking about it. Then it is excellent documentation "for the record". Code lives alongside documentation that describes it at a given point in time. Blog posts let you trace back where the decisions came from, and why. Why Writizzy went with feature flags, with a monorepo, why Writizzy picked MongoDB (and why Writizzy later decided to drop it). A post is history, not just a technical tutorial. Finally, it is a showcase for your brand. A blog reflects the culture of the team writing it. Plenty of candidates came to Writizzy through its posts, won over by the culture that showed through them, and they started the conversation right there. It is also an asset for your company's SEO. I have been talking about teams here, mostly engineering ones, but really anyone in your company can use this, along with everything else Writizzy brings: detailed analytics, newsletters tuned for deliverability, SEO audits and more. So of course, for Writizzy, I wanted team writing to feel first class: Writizzy comes with several roles, contributor, author and editor, and you can change them at any time. They define who can write, who can review and who can publish. Nice bonus: there is no limit on how many people can join a team. By default, everyone publishes their own posts, just like before. But if you would rather have a second pair of eyes on a post, the author sends it to review in one click. It lands in the "In review" tab, the reviewer gets an email, opens the post, leaves notes, and either publishes it or sends it back to the author. No more back and forth in Google Docs. Write and approve your posts in the same place, and leave review notes right inside the post to talk things through with your reviewers. You will never lose anything. Nothing is more frustrating than several people editing the same document and one of them silently overwriting the other. Writizzy now detects those conflicts, and keeps a version history so you can roll back easily. You should be building your own brand, not lending your reputation to someone else. It is better for your SEO, and it also makes it easy to move elsewhere later if you ever want to. And if you already write somewhere else, Medium, Substack, Ghost, WordPress or Beehiiv, your posts come over in a single import, images included. One last thing: send Writizzy your feedback. The product moves forward every day thanks to what its users tell Writizzy. Software Engineer with more than 20 years of experience. I love to share about technologies and startups About one email a month, when a piece is worth writing. Your readers stay your readers. 321 subscribers · unsubscribe in one click

Piper Sandler: 5 software stocks cutting AI token costs. By Peace Longe TheStreet Updated July 26, 2026 10:19 AM Gift Article For most of the past two years, investors bought nearly every company tied to semiconductors, from chip designers to equipment makers, and those stocks rose across the board. Enterprise software, on the other hand, got treated as collateral damage, priced as though large language models would eventually make the whole category redundant. That assumption is now getting tested, and not by the software companies themselves. Piper Sandler told clients on Wednesday that five infrastructure software names are positioned to solve the problem chief information officers complain about most: Running AI agents at scale costs far more than anyone budgeted. Piper Sandler's argument is that the customer data these companies already store can cut the number of tokens an AI agent needs to process, which lowers the cost of running it. Why Piper Sandler says these 5 software stocks cut AI token costs. The note, led by analyst Rob Owens, named Elastic (ESTC), GitLab (GTLB), MongoDB (MDB), Snowflake (SNOW), and Atlassian (TEAM) as the primary beneficiaries, Investing.com reported. A token is a chunk of text that is often smaller than a word. AI models charge by the token, counting both what you send in and what you get back. Owens wrote that the proprietary data already sitting inside these platforms can make models "significantly more accurate and efficient while dramatically reducing token usage costs." That will let companies expand AI adoption without costs rising too much. Early deployments showed token usage falling by 50% to 75% when clean organizational context was fed directly to the agent. The mechanism is simple enough. AI uses fewer tokens and answers faster when given clean, organized data instead of messy data. The token math that changed enterprise AI budgets in 2026. Here is the part that confused a lot of investors this year: Token prices fell, yet AI bills went up anyway. Owens noted that output tokens on newer frontier models run about 50% cheaper than the prior generation, yet improved reasoning capabilities caused consumption to increase. Reasoning models think in tokens, so a single query that once cost a few hundred tokens can now cost tens of thousands. More AI Stocks: Snowflake's pricing documentation shows how detailed this has become. The company splits AI usage onto a separate consumption meter so customers can track token spend against regular processing costs. That shift changed corporate behavior. Companies moved away from what Owens calls "Tokenmaxxing," or throwing unlimited model capacity at every problem. Instead, the companies shifted toward model routing, which sends easy queries to cheap models and hard ones to expensive models. What the consumption pricing model means for revenue. Vendors price context layers on consumption rather than per seat. That matters because the per-seat model is exactly what the market fears AI will destroy as headcounts shrink. Piper Sandler called this an attractive incremental growth opportunity that also strengthens long-term competitive advantages. Put plainly, if a customer's AI agents run more queries next quarter, the vendor gets paid more without signing a single new user. Three things have to hold for that thesis to work: * Enterprises must keep expanding agent deployments rather than pausing them. * Context layers must stay difficult enough to replicate that model vendors do not absorb the function. * Consumption revenue must grow faster than any decline in traditional seat licenses. Owens said conversations with management teams and channel partners confirmed that organizations are turning to software to make AI more efficient. How these 5 software stocks have actually traded. The stocks Owens named have not moved as a group. MongoDB has been the standout, with a market capitalization near $27.7 billion in mid-July, up more than 62% from last year, according to StockAnalysis data. The stock traded around $307 on July 21. Elastic went the other direction. Shares sat near $50 in recent trading, and Jefferies cut its target to $75 from $95 while keeping a Buy rating. GitLab has been the weakest of the five. Analysts carry an average Hold rating with a 12-month target of $34.50, roughly 4% above where shares trade. Snowflake sits in between, with 33 analysts rating it Strong Buy at an average target of $302.26. Atlassian rounds out the group with shares sitting near $86 as of the time of writing, well below the average analyst target of $139.70 reported on Yahoo Finance. KeyBanc set the most recent target at $115 on July 8 while keeping an Overweight rating, which points to about 33% above where the stock trades. Where this fits against the broader software selloff. Piper Sandler is not alone in making this argument. Morgan Stanley told clients this week that sentiment on software has become too negative, naming eight Overweight companies positioned for the AI era, Yahoo Finance reported. The firm raised a similar question: What happens to software growth once AI companies stop selling tokens below cost? The backdrop explains why these calls keep coming. The S&P 500 software industry index has fallen more than 25% from its October highs. The iShares Expanded Tech-Software Sector ETF (IGV) tells a similar story. It's down 13% this year. Meanwhile, the S&P 500 has gained close to 10% over the same stretch. Risks investors should weigh before buying the thesis. The counterargument to Owens' call is that AI model providers build retrieval and memory features directly into their own platforms. Nothing stops a frontier lab from building its own retrieval and memory tools, which would make a third-party context layer less necessary. Several labs have already started doing this. There is also a timing problem. Piper Sandler describes a critical window opening, which is analyst language for a call that has not yet shown up in reported revenue. None of these five companies breaks out context-layer revenue as its own line item in filings. That means investors are betting on an analyst estimate, not a disclosed number. Two further limits that also matter: * The 50% to 75% savings figure comes from early use cases, not audited results across a customer base * Consumption pricing cuts both ways, since AI budget cuts would hit revenue faster than annual seat contracts would What to watch next on these AI software stocks. The next earnings cycle should settle a lot of things. Snowflake, MongoDB, and Elastic all report consumption metrics that investors can check to verify Piper Sandler's call. Each company's management comments on AI-driven usage will tell you whether context layers are actually producing revenue. Watch net revenue retention specifically. If existing customers are spending more as agent deployments expand, that means the consumption approach is working. Also watch whether GitLab and Atlassian, the two most seat-dependent names on the list, can show credit or consumption revenue growing while seat counts stay flat. For readers deciding what to do with this, the practical read is that the five names carry very different risk profiles despite sharing a common call. MongoDB has already priced this call in. GitLab has not. The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc. This story was originally published July 26, 2026 at 10:07 AM.