Improving Research Outcomes

Anthropic is killing it with these technical posts. If you're an AI dev, stop what you are doing and go read this. It shows, in great detail, how to implement an effective multi-agent research system. Pay attention to these key parts: Anthropic shares how they built Claude's new multi-agent Research feature, an architecture where a lead Claude agent spawns and coordinates subagents to explore complex queries in parallel. They use the orchestrator-worker architecture. This system allows Claude to dynamically plan, search, and synthesize high-quality answers across large corpora using web, workspace, and custom tool integrations. Orchestrator-Worker Design The lead agent decomposes a query, spins up specialized subagents (each with their own tools, prompts, and memory), and integrates their results. This parallel, breadth-first design dramatically improves performance for research tasks over sequential LLM use. It yields 90% higher success rates in internal evals compared to single-agent Claude. Token-efficient Scaling Performance gains correlate strongly with token usage and parallel tool calls. By distributing work across multiple agents and context windows, Claude’s system scales reasoning capacity efficiently. However, this comes with a 15× token cost over standard chats, making it suitable for high-value queries only. Prompt engineering is not dead! Anthropic iteratively refined agent behavior via prompt design. They embedded heuristics for task complexity scaling, delegation clarity, tool selection, and thinking strategies. They also used Claude to self-optimize prompt and tool use, reducing task times by 40%. Flexible Evaluation + Production Reliability Anthropic uses LLM-as-judge scoring with rubrics for factuality, citation, and efficiency, alongside human testing to catch subtle failures. For reliability, they built resumable stateful agents with checkpointing, rainbow deployments, and full observability of agent decision traces, crucial for debugging non-deterministic, long-running agents.

At oncology conferences, we often clap when we see a “positive” clinical trial. A statistically significant p value. A hazard ratio below 1. A curve separating beautifully on a slide. And rightly so. These studies represent years of work, scientific innovation and hope. But afterwards, I often reflect on how differently patients interpret these results. Because patients are not asking: “What was the hazard ratio?” They are asking: “Will I live longer?” “Will I be well enough to work?” “Will I still be able to care for my family?” “Will this treatment make me sick?” “Is the benefit worth the side effects?” And this is where oncology becomes far more complex than statistics. Take a statement like: “18% reduction in breast cancer recurrence.” To many people, that sounds like: 18 out of 100 women avoided recurrence. But that is usually not what the statistic means. Often, this refers to a hazard ratio, which compares the relative risk between two groups over time. For example: A hazard ratio of 0.82 means the treatment reduced the relative risk of recurrence by 18% compared with the control group. But relative reductions can sound much larger than the actual benefit experienced by patients. Imagine: • Without treatment, 10 out of 100 patients develop recurrence • With treatment, 8 out of 100 patients develop recurrence The absolute benefit is 2%. So the treatment prevented recurrence in 2 extra patients out of every 100 treated. Suddenly, the conversation feels very different. Especially when treatment may also bring: • fatigue • diarrhoea • neuropathy • menopausal symptoms • cardiotoxicity • lung toxicity • fertility implications • financial toxicity • repeated hospital visits Some therapies are transformative and have changed the natural history of breast cancer but modern oncology is increasingly forcing us to think beyond statistical significance alone and consider: • magnitude of benefit • quality of life • toxicity burden • financial cost • patient priorities Because patients value outcomes differently. Some will accept significant toxicity for even a small reduction in recurrence risk. Others prioritise independence, work, family and preserving quality of life. Neither perspective is wrong. This is why communication matters. Not just presenting statistics. But translating them. Humanising them. Contextualising them. Because behind every hazard ratio is a human being trying to decide how they want to live THEIR life. With ASCO just days away and data coming fast, perhaps this is a gentle reminder for all of us. Not only to ask: “Did the trial meet its endpoint?” But also think, “How will we explain what this truly means to the patient sitting in front of us?” That is the real art of oncology. #DrBookiesNuggets #Oncology #CancerResearch #Statistics #ASCO2026 #PatientCentredCare #QualityOfLife #PrescribingStewardship

Science has long relied on single ML models - powerful, but limited because they are bound by baked-in knowledge. Our recent experiments show that genuine discovery may emerge when a very large number of agents interact, adapt, and co-create, much like biology itself. Last week at Harvard’s Big Data 2025 I shared how multi-agent swarms are potential paths to move beyond pattern-analysis to invent - exemplified in very difficult problem spaces such as de novo proteins and music with long-range form. The swarms are formulated like reinforcement-learning collectives, where agents learn on the fly, adapt to each other, and evolve strategies in real time, to yield hallmarks of intelligence that a single model cannot exhibit. Our swarm generated proteins well outside natural and single-model clusters (UMAP), and the music scored the highest small-worldness with the most long-range links, the signature of integrated, human-like creative structure. In a deeper analysis, when we mapped how themes in the music connect across time, the swarm built networks with the tightest balance of local clusters and global connections, beating all baselines. The resulting composition was not just repeating patterns stitched together, but showed organic global coherence without repetition, like sparks of creativity emerging on their own. Preprints coming soon!

I stopped obsessing over publishing… and that’s when my research career took off. (Here’s what I discovered) Four years ago, I believed publishing was the only path to academic success. My inbox? Empty. My collaborations? Stagnant. My impact? Limited to footnotes in other people’s papers. Then, I did something radical: I shared my work outside journals. A blog post about the LCA work I did for a partner. A LinkedIn post breaking down techno-economic assessment and process design methods. A webinar sharing my research outputs. Crickets. For weeks. Until a founder DM’ed: "Liked the recording of your webinar. Can you do something like this for us to verify our TEA?" A week later, an academic mentor slid into my DMs: "Saw your recent work on carbon capture. Can we co-write a research bid?" I wasn't sure what to do. This wasn’t “real” academic work. I’d been pre-conditioned to share my work only in scholarly journals and conferences. But suddenly, my research was solving problems, not merely gathering dust. So I leaned in. I built a simple system:  1. Every paper made available as PDF with posts on problem, method, outputs  2. Conference slides became PDFs shared with key takeaways  3. Complex science converted into trade magazines and blog posts The response? A FTSE 250 energy company invited me to perform a market study for their new direct air capture business strategy. Learned societies invited me as a keynote speaker for their events. And yes, citations keep coming, but now tied to real-world impact. Here’s what academia won’t tell you: Visibility isn’t vanity. It’s the bridge between your work and the problems it can solve. You don’t need 100 papers to make a difference. You need the right people to see your work. Now? I teach researchers to build this bridge. Because your career shouldn’t hinge on how many journals you’ve cracked. It should hinge on how many minds you’ve changed. Start there. Let the papers follow. P.S. What is the most significant challenge that pre ents you getting your research seen by others? #science #scientist #research #publishing #phd #postdoctoral #professor #academia #highereducation

I made this leadership mistake for 6 years. It cost me my best employees. When leadership is task-focused, teams stay stuck in execution mode. But when you shift from "do this task" to "achieve this goal," you unlock their true potential. Here’s what happens when you lead with outcomes instead of instructions: They stop following orders and start innovating. Ownership replaces dependency, and results skyrocket. 3 secrets to outcome-focused leadership: 1️⃣ Set crystal-clear goals: Everyone should know exactly what success looks like. 2️⃣ Provide total freedom: Trust your team to figure out how to achieve those goals. 3️⃣ Celebrate every win: Big or small, recognition fuels momentum. The magic is in ownership. When your team owns their outcomes, they don’t just complete tasks, they revolutionize. They write their own success stories and achieve results beyond your expectations. Trust is the foundation of breakthrough results. Give your team space, and watch them soar. How do you empower your team to achieve extraordinary outcomes? #leadership #teamgrowth #innovation

A Phase 3 trial can meet its endpoint and leave the market uncomfortable. That is probably the biggest takeaway for me from the LOTIS-5 readout with Zynlonta in relapsed/refractory DLBCL. The study was not a simple “failure.” Actually, from an efficacy perspective, the trial delivered. Zynlonta + rituximab improved progression-free survival versus R-GemOx, and other efficacy signals such as response rate, complete response rate, and duration of complete response also moved in the right direction. The real issue was safety. The experimental arm showed a higher rate of serious adverse events, more treatment discontinuations, and a relevant imbalance in Grade 5 treatment-emergent adverse events: 13.2% vs. 4.6% compared with the control arm. That is the kind of signal that immediately changes the conversation from “clinical benefit” to “benefit-risk.” And the market reaction reflected that concern, with the company’s stock falling around 52% after the announcement. To me, this is a very important case study in how investors interpret oncology data. A positive PFS result may support the scientific rationale, but if safety raises doubts about real-world usability, regulators and investors will ask a different set of questions: Who is really benefiting? Who is most exposed to risk? Can the safety signal be explained, managed, or isolated? And is the therapeutic window still acceptable? This is where the company’s next steps become critical. One possible way forward is a deep subgroup analysis. If the excess mortality is concentrated in a specific population, for example, patients ≥75 years, then the discussion may shift from “the drug is unsafe” to “the drug needs a more precise population For me, the broader lesson is this: In biotech, a catalyst event is not just about the headline result. It is about the quality of the total story. When a company is highly dependent on one asset, any uncertainty around safety can rapidly become a valuation event. Investors do not only price the data; they price the probability that the data can become an approvable, usable, and commercially meaningful product. So, although LOTIS-5 met its primary endpoint, the next chapter will depend on whether ADC Therapeutics can clearly explain the mortality imbalance, identify the patients with the best benefit-risk profile, and propose a credible regulatory and clinical management plan. In oncology, “positive trial” is not always the end of the story and Sometimes it is just the beginning of a much harder question: Positive for which patients and at what cost? I read you in the comments. If you want a deeper analysis, comment "ANALYSIS" and will send you the full report. #Biotech #Oncology #ClinicalTrials #DLBCL #ADC #DrugDevelopment #PrecisionMedicine #Investing

Are we measuring the wrong things in drug innovation? Some of the most valuable therapies might never show up on our innovation radar. The typical view in US #biopharma has long equated “innovation” with patents, new drug approvals, and R&D spend. They're easy to count and look good in investor decks. However, these metrics often reward volume more than total value. They don't tell us whether a therapy meaningfully improves patient lives, strengthens public health, or delivers returns beyond the financial metrics. A new six-dimensional framework published in The Incidental Economist offers another option. Drawing from over 600 interdisciplinary studies, the authors propose a more rigorous definition of #innovation: - Scientific and Technological Advances: Captures innovation and productivity using metrics such as new molecules, new drug applications, and patents. Emerging indicators, such as AI-enabled R&D and digital biomarkers, offer forward-looking insights. - Clinical Outcomes: Highlights therapeutic impact through metrics such as safety, efficacy, and patient-reported outcomes, emphasizing real-world patient benefits and delays in disease progression. - Operational Efficiency: Measures efficiency in development and production using trial success rates, R&D timelines, supply chain resilience, and adaptive trial designs. - Economic and Societal Impact: Evaluates economic returns and societal benefits through cost-effectiveness analyses, budget impacts, and productivity improvements. - Policy and Regulatory Effectiveness: Assesses how regulatory frameworks support innovation through approval speed, breakthrough designations, and surrogate endpoint integration. - Public Health and Accessibility: Examines broader health impacts, including reduced disease incidence, healthcare access improvements, and equitable geographic distribution, ensuring innovations meet widespread public health needs. This doesn't have to just be academic. It could change what gets funded, approved, and reimbursed. Some examples mentioned in the article: -An Alzheimer's therapy might look risky on paper, but when viewed through long-term productivity gains and reduced caregiver burden, it becomes a more attractive, high-risk/high-reward bet. -A platform technology (e.g., mRNA) may not boost new molecule counts today, but could enable faster, more precise drug development in the future. -A one-time gene therapy with high upfront cost could prove more valuable than chronic treatments when lifetime adherence and hospitalizations are factored in (if payers can afford the upfront investment). Of course, expanding how we define innovation introduces trade-offs. Complexity increases. Metrics will compete against each other. The question is whether the upside of greater alignment with ALL stakeholders is worth the operational complexity and potential reductions in value for some individual stakeholders. Would you be in favor of evaluating innovation more holistically?

You’ve probably run into this scenario in practice 👇 A patient on dialysis for chronic kidney disease who now needs chemotherapy This is one of those situations where things get complicated quickly Dialysis doesn’t just add a layer of logistics - it impacts outcomes. These patients tend to have higher mortality, both from cancer and other causes. And the hardest part is that we don’t have great data to guide us Most of what we know about chemotherapy use in patients on dialysis comes from case reports, small case series, or retrospective studies Even available guidelines can be inconsistent and often rely more on expert opinion than strong evidence On top of that, patients on dialysis are almost always excluded from clinical trials So we’re left piecing things together from limited data, variable experiences, and a lot of clinical judgment Managing these patients requires close coordination between oncology and nephrology Dialysis can remove certain chemotherapy drugs from the body, which directly affects how well they work So timing and dosing become critical Give a full dose of a renally cleared drug → higher risk of toxicity and worse outcomes Reduce the dose too much or have dialysis remove the drug too early → risk of undertreatment That balance is where most of the decision-making lives Chemotherapy is absolutely still possible in these patients But it usually requires: 👉 dose adjustments 👉 thoughtful scheduling (often after dialysis) 👉 and close monitoring When you’re approaching a patient on dialysis who needs chemotherapy, here are a few things to think through: 💡 Start with the goal Are you treating for cure or palliation? That should shape how aggressive you are with dosing and regimen selection. 💡 Understand the drug Is it dialyzable? Renally cleared? Are there alternative options that are less dependent on kidney function? Drugs that aren’t significantly removed by dialysis are often easier to work with. 💡 Consider the dialysis modality Not all dialysis are created equal - different modalities can impact drug clearance differently 💡 Coordinate the schedule Aligning chemotherapy timing with dialysis sessions is key to optimizing exposure 💡 Monitor closely These patients are more prone to complications like infections, sepsis, and anemia, so vigilance is critical Resources like the Onconephrotoxin Library Collaboration (OLIC) can be useful for drug specific insights. It’s a multidisciplinary effort (including pharmacists) that compiles and reviews the available literature to support decision-making in these complex cases. --- I’m the Kelley in KelleyCPharmD 👋 and I help pharmacists learn the complex world of oncology 📧 Want my help? Join 3000+ other pharmacists learning from my Oncology Insights Newsletter or DM me to learn about the ELO Collaborative, the only oncology pharmacist training program built with expert support, community, and real-world application at the center.

I found the missing piece for building AI agent teams that actually collaborate! Common Ground is an open-source framework for creating teams of AI agents that tackle complex research and analysis tasks through true collaboration. Think of it as simulating a real consulting team: a Partner agent handles user interaction, a Principal agent breaks down complex problems, and specialized Associate agents execute the work. Key Features: • Advanced multi-agent architecture with Partner-Principal-Associate roles • Full observability with real-time Flow, Kanban, and Timeline views • Model agnostic with built-in Gemini integration via LiteLLM • Extensible tooling through Model Context Protocol (MCP) • Built-in project management and auto-updating RAG system The breakthrough? It transforms you from a passive prompter into an active "pilot in the cockpit" with deep visibility into not just what agents are doing, but why they're doing it. Perfect for building agents that handle multi-step workflows and strategic collaboration beyond simple command-response chains. It's 100% open-source. Link to the repo in the comments! ___ Connect with me → Shubham Saboo I share daily AI tips and opensource tutorials on AI Agents, RAG and MCP.