The Daily

In biotech, "good science" is table stakes; what survives is what can stay alive in the clinic and in a budget. Today’s set is basically three versions of the same stress test: can the platform keep working when the easy part is over? In cell therapy, that means persistence (CAR-T that can renew itself) and immune evasion (islet cells that don’t trigger rejection without immunosuppression). In partnering, it means whether a discovery engine can keep producing assets that a big pharma partner actually wants to advance beyond preclinical. You can feel the market’s current bias: fewer "maybe" bets, more proof of repeatability, and more attention on manufacturing choices that change cell state rather than just add another construct. The next step to watch is who turns these results into protocol changes — i.e., what becomes a default manufacturing recipe, and what remains an interesting paper.

Astellas is exiting its long-running collaboration with CytomX on Probody T-cell engagers for solid tumors, effectively putting a stop to the remaining preclinical programs. The partnership started in 2020 with an $80M upfront and up to $1.6B in milestones — most of which now disappears from the table. CytomX had still been able to trigger milestones as recently as March/April 2024, when two bispecifics were nominated for further development, generating a combined $10M payment. But the latest update says Astellas chose not to advance the remaining assets. For CytomX, it’s another reminder that discovery platforms don’t get valued on optionality; they get valued on assets that move. The company says its "top priority" in 2026 is aligning with the FDA on a registrational path for its lead ADC varsetatug masetecan (mCRC), and it ended 2025 with $137.1M cash expected to last into Q2 2027.

Sana reported that its investigational allogeneic islet-cell therapy UP421 maintained function and secreted insulin for 14 months after transplantation in a patient with type 1 diabetes — notably without immunosuppressive medication. The investigator-sponsored first-in-human study (NCT06239636, Uppsala University Hospital) measured continued islet function via circulating C-peptide, including a spike after a mixed meal tolerance test (a practical "does it respond to food?" signal). The company also claims tighter glycemic control between months 12–14. The strategic point: UP421 uses donor-derived islet cells processed through Sana’s hypoimmune platform, aiming to avoid immune detection rather than brute-force suppress it. Sana says it will lean into its preclinical candidate SC451 (same hypoimmune platform, but stem cell–derived islet cells) with a Phase 1 planned as early as this year.

A team led by Albert Einstein College of Medicine describes a manufacturing tweak that aims directly at CAR-T’s persistence problem: instead of standard activation, they use a fusion scaffold protein HCW9206 that links IL-7, IL-15, and IL-21. The key result is a phenotype shift: more than half of the resulting CAR-T cells were T memory stem cells, compared with <5% under conventional manufacturing. In a mouse model of human leukemia, both approaches cleared tumor initially — but when the team simulated relapse by reinfusing leukemia cells weeks later, only the scaffold-generated CAR-T mounted a strong recall response and prevented recurrence. They also report improved performance in a humanized mouse model of HIV, including better elimination of infected cells. If this holds up in humans, it’s a rare kind of manufacturing change: one that modifies cell state (and potentially durability) rather than just optimizing yield or vector integration.

The interesting shift in CAR-T right now isn’t a new antigen — it’s the attempt to make outcomes predictable across centers, indications, and trial designs. That’s hard because the data is messy: tiny cohorts, inconsistent assays, and wildly different supportive-care baselines. But the newest work is increasingly "pan" in the way oncology got pan-genomic: aggregating enough patients and enough modalities (flow, qPCR, cytokines, clinical features) to build response signatures that aren’t just retrospective stories. In parallel, access and infrastructure questions keep surfacing: you can have a perfect biomarker and still not deliver therapy if the system can’t manufacture and deliver at scale. Today’s picks are a good pair: one paper about predictive modeling across trials, and one about the real-world map of CAR-T delivery. The next step is whether any of these signals becomes a decision rule that clinicians actually use pre-infusion.

This Nature paper takes a "pan-haematologic" approach to response prediction in CAR-T by analyzing 256 patients across 5 cancer types and 13 clinical trials. The dataset is unusually multi-layered: pre-infusion clinical variables, >2 million apheresis T cells profiled by flow cytometry (17 markers), ex vivo expansion metrics during manufacturing, >90,000 measurements of 30 serum markers, and serial CAR-T tracking via qPCR. The authors explicitly call out the practical barrier to applying ML in CAR-T (small sample sizes and non-uniform data generation), and then basically try to brute-force it with breadth. The takeaway isn’t a single magic marker; it’s the demonstration that some features generalize across diseases and trials — which is exactly what you need if you want a usable pre-infusion risk/response model.

This paper zooms out from molecules to the delivery system: CAR-T outcomes depend on referral speed, center capacity, and the realities of national reimbursement. The authors map European access patterns and, in doing so, underline an uncomfortable truth: "available" is not the same as "reachable" when the therapy requires specialized centers, ICU-ready supportive care, and tight logistics. Even as more indications open up, the bottleneck becomes operational — not scientific — and that means inequities can widen if capacity doesn’t scale alongside approvals. The practical implication is that access metrics should be treated like clinical endpoints: if you can’t get patients to therapy in time, response rates become a misleading comfort.

Solid tumors remain the "hard mode" for CAR-T, and one repeat offender is the microenvironment — hypoxia, suppressive cytokines, and dysfunctional vasculature. This study explores an "armored" CAR-T concept by engineering cells to secrete a VEGF-neutralizing scFv, aiming to locally counter pro-angiogenic, immunosuppressive signaling. The point isn’t that VEGF blockade is new — it’s that putting it into the CAR-T payload is a way to make the cell therapy itself a microenvironment-modifying agent, potentially reducing dependence on systemic combination dosing. The next step to watch is safety/PK logic: localized secretion is attractive, but any meaningful VEGF modulation can still have vascular side effects. Still, as a design pattern, microenvironment payloads are becoming the more serious "second act" for CAR-T.

The last year of AI was dominated by model releases; the next year looks like distribution and control systems. Shopping agents are a distribution war in disguise: whoever owns the "front door" to intent will tax everyone else. In enterprise software, the bottleneck is already shifting to process: the limiting reagent isn’t code generation, it’s review and governance, because nobody ships without trust. And in defense procurement, AI is increasingly bundled as "commercial solutions" that collapse hardware + software + services into one contract vehicle — which favors a handful of fast-moving vendors. All three are the same story: agents don’t just answer questions, they touch systems, and the fight is over who gets to authorize that touch.

Shopify president Harley Finkelstein argues that AI shopping agents will become personal shoppers and expand e-commerce beyond today’s penetration (he cites ~18% of U.S. retail purchases being online). The premise is that agentic shopping adds context and memory: if an agent learns your preferences, it can bias discovery toward what you actually like rather than what’s best at buying ads. He positions this as a discovery unlock for smaller Shopify merchants who currently struggle to get found. Shopify is also building merchant-facing agents like Sidekick, plus an agent for support workflows, and a protocol so agents can understand merchant product data. The near-term reality will be messy — search already personalizes — but the strategic read is right: if the interface shifts from "search results" to "delegated purchase," every retailer needs an agent strategy.

Anthropic launched Code Review, an AI reviewer integrated with GitHub that analyzes pull requests and leaves comments directly on code, focusing on logical errors over style. The pitch is straightforward: "vibe coding" increases output, but it also increases bugs and security risks — and it floods teams with more PRs than humans can realistically review. Anthropic says the tool runs in a multi-agent setup, where multiple agents inspect the code from different angles and a final agent aggregates and prioritizes findings. It labels severity with colors (red/yellow/purple) and targets enterprise customers first (Claude for Teams/Enterprise) in a research preview. A hard detail worth anchoring on: Anthropic estimates $15–$25 per review on average, token-priced — i.e., governance becomes a line item.

AMI Labs, co-founded by Yann LeCun after leaving Meta, raised $1.03B at a $3.5B pre-money valuation to build "world models" — systems that learn from reality rather than just language. CEO Alexandre LeBrun frames it as fundamental research that may take years to become commercial, explicitly contrasting it with typical applied AI startups that ship revenue in months. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with a long list of strategic/angel backers; AMI says it will publish papers and open-source code as it goes. The first disclosed partner is the digital health startup Nabla, which LeBrun ties to the limits of LLM hallucinations in healthcare settings. The meta-signal: "world models" is becoming a capital category, not just a research phrase.

The U.S. Army announced a 10-year contract with Anduril that could be worth up to $20B, structured as a five-year base period plus an optional five-year extension. The deal covers hardware, software, infrastructure, and services, and the Army describes it as a single enterprise contract consolidating more than 120 separate procurement actions for Anduril’s commercial solutions. It’s procurement logic that matches modern software: fewer bespoke buys, more platform-style contracting that can ship faster. For the broader AI market, it’s also a reminder that "agentic" and autonomy aren’t confined to office workflows; they’re becoming part of defense IT modernization — with all the policy and supply-chain fights that follow.

March in Vienna is usually the in-between month: not quite spring, no Christmas lights, and everyone is mildly tired of indoors. The workaround is to pick events that feel like a change of scene without demanding a whole day of effort. Immersive exhibitions are popular for exactly that reason — they’re weather-proof, they work on weekday evenings, and they give you a "trip" without leaving the city. In parallel, Vienna has one of Europe’s best low-friction art rituals: pop-in, browse, leave with something in hand — no gatekeeping required. Today’s two picks are built for that reality: one big, cinematic evening slot (Marx Halle), and one casual weekend wander (Mariahilfer Straße) with prices designed not to insult normal people. If you want a simple next step: pick the one that matches your energy level.

The Titanic gets the full immersive treatment at Marx Halle: augmented reality, virtual reality, music and theatrical staging to walk visitors from construction to maiden voyage to sinking. Vienna.info frames it as a tech-heavy exhibition with "impressive visual effects" — basically the blockbuster format that works well when the weather is undecided. It’s also a good group pick: you don’t need niche knowledge, and it scales from date night to family outing. If you go, the practical move is to treat it like cinema: book a defined slot, and then attach dinner/drinks around it. (Marx Halle’s location makes that easy.)

The "Supermarket of Art" concept is exactly what it sounds like: affordable fixed prices, a huge wall of originals, and zero intimidation. This season runs until March 28, 2026 on the Mariahilfer Straße, with around 95 artists from 14 countries showing more than 5,000 works (drawings, watercolor, acrylic/oil, photo, small sculptures). Entry is free, which makes it an easy "walk in, walk out" plan for a Saturday. It’s also one of the better ways to buy art in Vienna without instantly landing in a price bracket that makes you regret having a bank account.

Diese Woche ist das klassische "vielleicht ja, vielleicht nein"-Wetter — und mit Kindern heißt das: Plan A und Plan B müssen beide funktionieren, ohne dass man komplett umdisponiert. Mein Rezept: zwei Indoor-Anker (die man notfalls spontan macht) + ein Draußen-Teil, der bei Sonne sofort besser wird. Für März ist außerdem alles Gold, was eine klare Saisonmarke hat: Ostermärkte starten, Parks werden wieder nutzbar, und plötzlich gehen kurze Nachmittage wieder als Ausflug durch. Unten deshalb drei sehr pragmatische Bausteine: ein Monats-Programmüberblick, ein konkret planbarer Ostermarkt-Guide, und ein "Frühling in Wien"-Spickzettel für Tage, an denen ihr nicht lange diskutieren wollt.

1000things sammelt für März eine breite Vorschau — praktisch, wenn man nicht jeden Tag fünf Event-Kalender scannen will. Der Wert ist weniger "News" als Struktur: ihr seht auf einen Blick, welche Wochenenden thematisch voll sind (Märkte, Ausstellungen, Festivals) und wo die Stadt eher leer wirkt. Für Familien ist das hilfreich, weil man Termine früh fixieren kann, bevor alles voll ist. Nutzt es wie einen Baukasten: 1 Indoor, 1 Draußen, 1 Snack-Stop. Und ja: auch wenn es banal klingt, ein "wir machen heute nur ein kleines Ding" ist oft der beste Plan.

Ostermärkte sind im März/April das beste "low effort, high reward"-Format mit Kindern: draußen, aber strukturiert; Essen, aber nicht gleich Restaurant; genug zu schauen, ohne dass man lange laufen muss. 1000things listet die wichtigsten Märkte in Wien und gibt euch damit ein klares Raster: Welche Märkte sind zentral (Freyung/Innenstadt), welche sind "Ausflug" (Schönbrunn), und wo gibt’s typischerweise das meiste Kinderprogramm. Der Trick ist Timing: geht lieber früher am Tag (und unter der Woche, wenn’s geht), dann ist es entspannter. Und nehmt euch einen Markt als Fixpunkt pro Woche — sonst fühlt es sich schnell nach "wir waren eh überall" an, ohne dass es hängen bleibt.

Diese Liste ist der praktische Notfall-Knopf für Wochen, in denen euch die Ideen ausgehen: Ausflüge, Parks, Museen, Tier- und Natur-Optionen. Für März ist das besonders nützlich, weil draußen schon wieder geht, aber nicht zuverlässig. Der Mehrwert ist die Bandbreite: von kurzen Nachmittags-Plänen bis zu halben Tagen. Macht daraus eure eigene Shortlist (3–5 Favoriten), dann müsst ihr nicht jedes Mal neu entscheiden.

If you want Italy to feel like "Italy" (and not like a queue management system), you travel for the calendar: spring events, local food rituals, and low-season museum days. Venice is at its best when you have a reason to be there that isn’t just "I should see it once" — because then you naturally end up in neighborhoods at the right time and you don’t fight the midday tide. Puglia is similar: the region’s most memorable experiences are often local traditions that look simple on paper but feel deeply specific when you’re there. Today’s three picks are all calendar anchors: a museum-access day in Venice, a spring festival at Forte Marghera, and a San Giuseppe food tradition in Salento that is basically culinary anthropology with a schedule.

Venezia Unica is promoting "Musei in Festa" as a way to structure a Venice visit around museum access rather than peak-hour landmarks. These types of city-wide museum initiatives are underrated: they give you a reason to build a route (Dorsoduro → San Polo → Castello) that naturally avoids the worst crowd funnels. If you’re going with kids, it’s even better — museums become "stations" with indoor breaks and predictable timing. The next step is simply operational: pick the date, preselect 2–3 museums maximum, and book lunch early. Venice punishes improvisation.

Forte Marghera is one of those places that makes Venice feel larger than the islands: more space, less postcard pressure, and a "local weekend" vibe. Forte in Fiore is positioned as a spring event with the typical ingredients (flowers, stalls, outdoor programming) — and the important detail is the date: April 19, 2026. If you’re planning an Easter-ish trip, this is a clean anchor to decide which weekend you want. Pairing it with a quiet Venice day before/after can be a great two-speed itinerary.

In Salento, "Tavole di San Giuseppe" is a living tradition: an enormous ritual meal built around community hospitality. Visit Puglia describes the Minervino di Lecce version as running 17–19 March with a table featuring 169 dishes — an almost absurd level of specificity that signals how codified the tradition is. It’s not fine dining; it’s cultural memory made edible. If you’re anywhere near Lecce around these dates, this is the kind of event you build a trip around: you get food, music, and a social setting that tourists rarely stumble into by accident. The next step is to check local logistics (where/when the tables are set, and how access works), because these events are about community flow, not ticketing.

This Puglia.com piece is useful as context: it frames the "Tavolate" as a mix of folklore and solidarity, not just a food fair. The tradition is tied to offering and care — which is why the dish count and the ritual setup matters so much. For travelers, that’s the lesson: show up respectfully, don’t treat it like a buffet challenge, and you’ll likely have a deeper experience than any beach day. It’s also a reminder that Puglia’s best "events" are often not festivals with sponsorship banners, but older structures that locals still run like a duty.

By mid-March, teams are either refining playoff habits or quietly stress-testing young lineups and emergency rotations. The most telling moments aren’t the first-quarter runs; they’re the late-game possessions where tired legs and scouting collide. That’s why fourth-quarter takeovers matter: they’re not just "hot shooting," they’re a signal that a team can manufacture offense when the other side knows exactly what’s coming. Injuries and rest also reshape narratives: a contender’s "maintenance" can be a tanking team’s opportunity to build confidence and reps. Today’s games were good examples of both: Boston closed with defense and a finishing run, while New Orleans handled a depleted Dallas — but still got meaningful production from supporting pieces.

Boston beat Phoenix 120–112 behind a monster night from Jaylen Brown, who scored 41 with 18 coming in the fourth quarter. Jayson Tatum added 21 points in his fifth game back from an Achilles injury, while Derrick White (fresh off February East Defensive Player of the Month) had 21 and Payton Pritchard 19. The game turned late: Boston trailed 111–108 with under four minutes left, then Brown stole the ball from Devin Booker, fed Tatum for a layup, and the Celtics finished with an 8–0 run and 12 of the last 13 points. Phoenix wasted 40 from Booker and dropped a second straight after a brief win streak.

New Orleans handled an injury-depleted Dallas team 129–111 with Zion Williamson scoring 27 on an ultra-efficient 11-of-13 shooting in 28 minutes. Saddiq Bey had 23, with Jeremiah Fears and Trey Murphy III adding 17 each, as the Pelicans won for the 8th time in 12 games. For Dallas, Naji Marshall scored 32 against his former team, and rookie top pick Cooper Flagg had 21 points, 8 assists, and 7 rebounds. The Mavericks were missing Klay Thompson (rest), Daniel Gafford (illness), and Caleb Martin (sore foot), while New Orleans played without Dejounte Murray (illness) and still got a big rim-protection night from Yves Missi (10 rebounds, 5 blocks). This is the kind of March game that’s less about standings and more about who's still functional with the second unit.