Anglais

Google est fin février 2026 lancement de Nano Banana 2, également décrit par Google et une certaine couverture comme Gemini 3.1 Flash Image, peut ressembler à une mise à jour de modèle de routine, mais dans les cliniques et les groupes de recherche il se comporte plus comme un changement de flux de travail. Il se situe au milieu de pressions de production familières : matériel de recrutement limité par les délais de la CISR, chiffres de conférence qui doivent rester lisibles à distance, infographies patientes qui simplifient sans déformer, et l'enseignement de diapositives qui communiquent le risque de façon responsable. L'allégation de base de Google est qu'il associe la haute qualité d'image avec la vitesse de niveau Flash, maintenant accessible via l'application Gemini et les expériences de recherche Google tels que le mode AI et Lens, avec la messagerie de déploiement explicite pour MENA. Lorsque la génération d'images devient à la fois rapide et visuellement persuasive, le changement n'est pas seulement une sortie plus élevée, c'est des cycles de décision plus rapides. Google met en évidence les caractéristiques qui s'affichent directement sur la communication scientifique et clinique : amélioration du rendu des textes à l'intérieur des images, plus grande cohérence des sujets entre les éléments, contrôle de l'aspect-ratio et sorties jusqu'à 4K. Dans la pratique, cela peut signifier moins d'itérations «de conception manuelle» pour les visuels de titrage des médicaments, les diagrammes de processus de consentement, les résumés visuels et les schémas explicatifs, en particulier dans les équipes sans support de conception dédié. Une affirmation particulièrement conséquente est que Nano Banana 2 peut puiser sur des informations et des images en temps réel de la recherche Web pour améliorer la précision et le support des diagrammes et infographies. L'avantage est immédiat: des visuels plus clairs produits rapidement, susceptibles d'améliorer la compréhension du patient et de réduire la charge cognitive pour la formation

Clinical and research work rarely fails because we cannot locate information; it fails because we cannot convert information into usable outputs quickly enough. The friction lives in the in‑between tasks: preparing a background brief for a case conference, turning scattered notes into a coherent protocol amendment, cleaning citations before a resubmission, or drafting a patient-facing handout that is both readable and accurate. What is newly compelling about the idea of an “AI computer” is that it aims at this connective tissue of work, not simply at conversation. Perplexity describes Computer as a unified, cloud-based system that can “research, design, code, deploy, and manage” projects end-to-end, breaking a goal into sub-tasks and routing them across specialized components. In the public description, it can orchestrate work across 19 models in parallel, “matching each task to the best model,” while also remembering prior context and connecting to external services. The ambition is not subtle: it is a bid to turn the browser into an operational workspace where the user specifies intent and the system performs the multi-step work. The attraction is easy to understand in real workflows. We already practice constrained delegation every day: we assign parts of a project to trainees, research assistants, administrators, or colleagues, and we integrate, verify, and sign off. An agentic system promises a similar pattern, but at a different speed and scale. If it performs reliably, it may create time for what remains stubbornly human: therapeutic presence, clinical judgment under uncertainty, nuanced supervision, and careful interpretation of evidence. That tension becomes sharper when orchestration is invisible. If Computer decomposes a task into steps, chooses tools, and merges results, then provenance matters: which sources were used, which model generated which claim, and what was the system’s uncertainty at each step? In research, these details determine whether a literature synthesis is reproducible; in clinical settings, they determine whether a handout, policy memo, or documentation aid stays within the boundaries of evidence-based practice. The more autonomous the workflow, the more we need systems that make their reasoning legible rather than merely impressive. A second tension is provenance. In research, we need to know what sources were used, how claims were derived, and what uncertainty remains, because the credibility of a synthesis depends on traceable reasoning. In clinical environments, provenance is equally important, though less often formalized: we need to know whether an output is grounded in guidelines, high-quality trials, local policy, or merely plausible generalizations. Agentic tools can compress steps so efficiently that they also compress our visibility into where a conclusion came from. Cost pulls these questions from theory into daily decision-making. At roughly $240 per month, this is not an impulsive subscription for most clinicians; it is closer to a staffing tradeoff. Paying that amount implicitly assumes that time saved is both substantial and dependable, and that the time we spend verifying the output does not quietly re-inflate the workload. In clinical settings, the “true cost” includes not only money, but also the cognitive burden of oversight and the reputational risk of errors. From a practice perspective, the safest near-term uses are those that keep identifiable data out of the system and keep verification firmly human. Drafting non-identifying psychoeducation templates, creating training materials for interns, turning internal procedures into clearer language, or generating first-pass outlines for research documents can be sensible, provided we treat outputs as drafts and insist on source checking. The risk profile changes sharply when we move toward identifiable case details or highly specific clinical recommendations, especially in small communities or rare presentations where re-identification can be easier than we like to admit. We also need to acknowledge a quieter limitation that experienced researchers recognize: these tools can accelerate the appearance of scholarship. They can produce coherent framing, persuasive prose, and tidy synthesis even when the evidence base is mixed or contested. The danger, then, is not only “hallucination” in the headline sense; it is routine overconfidence, particularly under deadline pressure, fatigue, or institutional incentives that reward speed over carefulness. Ethically, we should treat agentic systems as a new layer of professional delegation that demands transparency and documentation habits. If AI materially shaped an output that informs care (a clinic policy, a patient handout, a decision support memo), the clinician’s responsibility is not reduced; it is reconfigured. We owe patients and colleagues a disciplined stance on what data entered the system, what sources were relied upon, and how claims were checked. This is consistent with broader AI-risk frameworks emphasizing lifecycle governance: mapping likely failures, setting boundaries, and building habits of verification rather than relying on good intentions. Looking ahead, the question is not whether “AI computers” will become more common; they likely will. The more important question is whether they become legible: systems that make their sources, assumptions, and limitations visible enough for clinical and research cultures that depend on auditability and trust. If we adopt them thoughtfully, starting with low-risk tasks, measuring time saved against time spent verifying, and maintaining strict boundaries around sensitive data, we can treat these tools as assistants rather than authorities, and preserve the integrity of our work while reducing avoidable friction.

As AI systems enter clinical workflows, terminology needs to become operational rather than rhetorical. An AI agent refers to a model-based software system capable of planning and executing multi-step tasks toward a goal, including selecting actions, using external tools, and revising steps based on outcomes. In other words, it’s more than a chatbot reply: it’s a goal-directed workflow runner. This definition matters because a lot of confusion starts when we treat any fluent system as “agentic.” Google DeepMind just dropped a long, ~42-page warning about why most “AI agencies” will fail in the real world, and when we read Intelligent AI Delegation (published February 12, 2026), it quietly explains the same thing from first principles. The uncomfortable takeaway is simple: a lot of what’s marketed as “agents” today won’t survive contact with reality. Not because they can’t write or plan, but because they can’t delegate in the way real systems require. And therapy is exactly where that gap becomes visible early. Here’s the part that stings: most agents today aren’t really agents in the strong sense people imagine. They’re task runners with good branding, we give them a goal, they break it into steps, call tools, and return an output. That can be powerful automation, and it can absolutely create efficiency in low-stakes work. But it’s not delegation, it’s a prettier pipeline. And in healthcare, “pretty pipelines” are exactly how hidden risk slips into daily practice. So what do we mean when we say “delegation”? Delegation isn’t one thing. It can mean delegating execution (the system carries out steps you already chose), delegating workflow control (it sequences tools, retries, and manages handoffs), or delegating judgment (it decides what should be done and when, based on interpretation of context and risk). Most systems marketed as “agentic” are mostly capable of execution and some workflow control, not delegation where judgment is offloaded and accountability meaningfully shifts. In therapy, the danger is when we unintentionally delegate judgment while assuming we only delegated execution. DeepMind’s brutal point is that real delegation isn’t just splitting tasks. Delegation is transferring authority, responsibility, accountability, and trust, and doing it dynamically as the situation changes. That means we don’t only ask, “Who can do this fastest?” We ask, “Who should be trusted to do this, under these constraints, with these consequences?” Almost no current system behaves that way end-to-end once multiple tools, uncertain data, and real stakes are involved. Before delegation even happens, the framework implies we need to evaluate capability, risk, cost, verifiability, and reversibility. In other words, it’s not “does the agent have access to the calendar tool?” It’s “is this task safe to hand off, can we verify the result, and can we undo harm if it’s wrong?” That is a very clinical way of thinking, and it’s exactly why this paper hits therapists harder than it hits casual tech demos. Our work already assumes risk, not perfection. Therapy is not just information work; it’s relational work under ethics, confidentiality, and duty-of-care. Our clients don’t experience our work as “outputs”, they experience it as trust, safety, timing, attunement, and boundaries. So when an AI system is framed as an “agent,” the real question becomes: what are we delegating, and what are we accidentally outsourcing? If we let a system drift into clinical judgment, we may be delegating more than we think. When AI agency fails in our context, it often fails without obvious nonsense. It fails quietly: a note becomes subtly distorted, a risk signal gets downplayed, a client message becomes overly confident, or a safety-plan phrase becomes too generic to be safe. The output can look coherent and professional, sometimes more professional than we would write at 6 p.m. and that’s precisely what makes it dangerous. Fluency can create false reassurance, and false reassurance is a clinical risk. Current limitations of AI agents (especially in real workflows) Even when agents perform specific tasks well, they often struggle with coordination and interoperability, working reliably across tools, policies, datasets, and other agents. Multi-agent setups can become brittle: one agent’s assumption becomes another agent’s “fact,” tool errors propagate, and responsibility becomes hard to trace. This matters because real clinics aren’t single-tool sandboxes; they’re multi-system environments with partial data and shifting constraints. The limitation isn’t just intelligence, t’s dependable, governable collaboration. Because of limited interoperability, humans are still required to manage coordination between multiple agents. This is the governance problem: accountability diffusion. If one agent delegates to another, which calls a tool, which references incomplete data, who is actually accountable for the final wording sent to a client? We might sign it, the clinic might deploy it, the vendor might supply it, and the model might generate it, yet no one can point to a clear, auditable decision boundary. In therapy, that’s not abstract; it’s how ethical responsibility gets blurred. Still, we shouldn’t throw the tools away, and we should keep the narrative balanced. Agents promise real efficiency: they can reduce admin burden, draft psychoeducation we approve, structure worksheets, translate or simplify content to match literacy needs, and summarize our own prior notes with clear provenance. But practical constraints remain, so boundaries must stay explicit: delegate low-risk, reversible, verifiable tasks, and resist delegating anything that functions like diagnosis, risk assessment, crisis response, or treatment decisions. If we get that boundary right, we take the gains without surrendering what clients come for: trustworthy, human accountability.

In the past two weeks, many clinicians have seen posts about Lotus Health AI, described as an “AI doctor powered by real doctors”, a physician-supervised service that can assess, diagnose, prescribe, and refer, positioned as free, always available, and multilingual. The company also announced a $35M Series A bringing total funding to $41M, co-led by Kleiner Perkins and CRV, and argues it can cut administrative waste by making physicians more productive without insurance billing. From a therapist’s perspective, the key question isn’t whether the marketing is compelling, it’s whether this becomes a reliable front door to primary care that reduces the friction therapy often absorbs: sleep problems mistaken for depression, thyroid/anemia issues that look like burnout, medication side effects that destabilize mood, and medical uncertainty that fuels panic and health anxiety. In day-to-day work, therapy can become the place where fragmented healthcare gets processed, emotionally, and sometimes administratively, because clients can’t access timely care elsewhere. One practical benefit, if Lotus functions as described, is speed as clinical timing, not just convenience.Earlier medical clarification can make therapy interventions more accurate and effective.When a client says, “Something feels off physically,” we validate and explore patterns, yet sometimes the responsible step is: get a medical assessment early.The differential matters: if symptoms are physiological or medication-related, the plan shifts; if they’re anxiety-driven, it shifts differently.In many systems, “soon” becomes weeks, and therapy becomes a holding space for unresolved uncertainty.A 24/7 primary-care-like channel that reviews consolidated history and routes to in-person care could reduce that gap.Lotus explicitly positions itself as going beyond generic chatbot advice by involving board-certified physicians who review guidance and prescribe when appropriate. A second benefit is indirect but meaningful: it may improve the quality and coherence of information clients bring back into therapy. Many clients struggle to describe symptoms clearly due to stress, dissociation, executive dysfunction, trauma, or exhaustion. If a platform helps organize meds, labs, and prior records into a clearer story, it can reduce shame (“I’m not making it up”), sharpen insight (“my panic clusters around sleep disruption”), and help therapists choose more precise interventions (e.g., exposure for health anxiety vs urgent referral for red flags). Third, there are coordination and access benefits even if therapists never touch the platform. When clients can obtain refills, medication reviews, and referrals with fewer obstacles, therapy is less likely to be derailed by preventable destabilizers. Practically, that can mean fewer sessions spent troubleshooting access and more time spent on the core work of therapy, skills, meaning-making, attachment repair, behavior change, and identity-level integration. That said, therapists are trained to notice how tools can become part of a symptom cycle. An always-available “doctor in your pocket” may stabilize some clients, but it can also feed reassurance-seeking for others, especially with health anxiety, OCD-spectrum checking, panic, somatic symptom patterns, or trauma-related body scanning. Even if guidance is solid, repeated checking can still function as avoidance or compulsion. The goal isn’t to demonize the tool, but to integrate it into a treatment plan with clear, time-limited, values-consistent use. There are also boundary questions that show up quickly. If clients rely on app-mediated care, therapists will be asked to interpret it: “Should I trust this?” “What does this diagnosis mean?” “Can you message them for me?” A helpful posture is to treat it like any outside provider: help clients clarify questions, process impact, and decide next steps, while avoiding medicine-by-proxy. It also matters that what any physician-led digital service can do may vary by jurisdiction, licensure, and telemedicine rules, so “available” may not always mean “authorized to treat” in the way clients assume. Ethically, the business model matters because trust is a clinical ingredient. If clients suspect recommendations are influenced by sponsorships or commercial incentives (whether or not that’s true), it can erode trust in healthcare and show up in therapy as cynicism, avoidance, or hopelessness. This makes transparency practical, not philosophical: what is automated vs clinician-reviewed, and how conflicts of interest are managed over time. Privacy and data integrity are equally central. Even with strong security claims, it helps to think concretely: what data is being linked, who can access it, what consent was given, what can be deleted, and what might be retained. Data-rich systems can fail in new ways, through breaches, misuse, or overconfidence in incomplete records, so the most ethical stance is careful integration, explicit consent, and humility. Overall, I’m cautiously interested in what this could become for therapy-adjacent care: faster evaluation, less avoidable suffering, and less informal care coordination falling onto therapists. But the promise depends on whether it earns trust in real-world use, clinically, ethically, and operationally, so clients aren’t left alone with medical uncertainty that insight alone can’t resolve.

AI in healthcare is no longer a concept. It’s a reality in motion and we’re now watching it mature in a more serious, more structured way. For a long time, healthcare was treated like a “possible use case” for general AI. But in practice, it has already been one of the most common real-world arenas where people test what these systems can do. Patients ask questions after clinic hours because anxiety doesn’t follow office schedules. Caregivers try to interpret lab results while waiting for a follow-up appointment. People navigating chronic conditions search for plain-language explanations of complex treatment plans. Clinicians, meanwhile, are under relentless documentation pressure and are constantly looking for tools that reduce cognitive load without sacrificing safety. The demand wasn’t hypothetical, it was already here. What’s changing now is that major AI companies are building healthcare-specific products that admit, plainly, that medicine is not “one size fits all.” Two launches make the shift easy to see: Claude for Healthcare and ChatGPT Health. They’re often discussed in the same breath, but clinically and ethically, we shouldn’t treat them as interchangeable. They point to two different problems and two different audiences, and that distinction matters because it shapes how risk shows up. ChatGPT Health is best understood as a patient-oriented space: a place where individuals can connect personal health or wellness information and receive explanations, summaries, and context in language that feels human. The promise is clarity. Healthcare is full of jargon, fragmented portals, and rushed appointments; a tool that helps someone understand their own information could reduce confusion and improve follow-through. When used appropriately, it can support better conversations with clinicians because patients arrive with sharper questions and less overwhelm. But that same strength is also its most predictable risk. When a system explains something smoothly, people can mistake fluency for clinical authority. We’ve all seen it: a confident tone can feel like certainty, even when the underlying situation is ambiguous. In healthcare, that gap is not academic, it can shape real decisions. So the safety challenge for a patient-facing tool isn’t just accuracy in a narrow sense. It’s expectation-setting, clear boundaries, and guardrails that prevent “informational support” from being interpreted as diagnosis or medical instruction. Claude for Healthcare, on the other hand, is more naturally framed as an enterprise and workflow tool. The emphasis is less “ask me anything” and more “connect me to the work.” Healthcare organizations don’t just need answers; they need operational support: interpreting and summarizing complex information at scale, reducing administrative friction, supporting research and internal processes, and fitting into existing systems without turning every task into another tab and another login. If we’re honest about what burns clinicians out, a large portion of it lives here, in documentation, administrative tasks, and the endless effort of finding and re-finding information across messy workflows. That’s why workflow-oriented tools feel so attractive: they target the pressure points that are actually breaking the system. But again, the risk is different. When AI becomes part of a workflow, it can scale both efficiency and error. If an output is wrong and nobody catches it, the mistake doesn’t just affect one conversation it can become embedded in documentation, passed forward, copied, and normalized. The more “plugged in” the tool is, the more essential it becomes to design for oversight rather than speed alone. This is where the compliance conversation becomes more than a checkbox. Once these systems touch sensitive health information, the question isn’t simply “is it HIPAA -compliant?” or “is it GDPR-compliant?” The deeper question is: how will data governance work in real life, with real people, under real time pressure? HIPAA (Health Insurance Portability and Accountability Act) and GDPR (General Data Protection Regulation) are different frameworks, but both push toward the same discipline: clear rules about what data is collected, why it’s collected, who can access it, how long it’s kept, and what happens when something goes wrong. And here’s the point we can’t afford to gloss over: compliance is not something a company can fully “grant” to a user through a product announcement. Even if a system is designed to be HIPAA-ready or offers strong security features, organizations still need to deploy it responsibly. That means access controls, role-based permissions, audit trails, staff training, retention policies, incident response planning, and crystal-clear boundaries around what data should and should not be entered into the system. For patient-facing AI like ChatGPT Health, privacy and consent have to be especially legible, because patients don’t always realize what they’re sharing when they upload documents, connect accounts, or paste text from portals. The tool needs to prevent accidental oversharing and make it obvious when a question crosses into “you need a clinician” territory. For workflow-oriented AI like Claude for Healthcare, the burden shifts toward institutional controls: connector permissions, least-privilege access, monitoring, and accountability structures that keep “helpful automation” from becoming invisible decision-making. What’s genuinely promising in all of this is the direction of travel. We’re moving away from the idea that one generic assistant can safely serve every healthcare scenario. We’re seeing specialization: tools designed for patient understanding, and tools designed for clinical and organizational workflows. That specialization makes it easier to define what the system is for, how it should be evaluated, and where the boundaries must be enforced. Our take is simple: we’re watching healthcare AI diversify, and that’s a sign the sector is being taken seriously. But seriousness comes with obligations. These tools should assist, not diagnose. They should reduce burden, not quietly introduce new error channels. And they should handle sensitive data with governance that is operational, not rhetorical. If we build with those principles, AI can improve clarity and relieve real pressure in healthcare. If we don’t, it will scale confident-sounding uncertainty into the one domain where people can least afford it.

As researchers, we spend a surprising amount of time doing work that is not, strictly speaking, research. We design studies, refine hypotheses, collect and analyze data, and engage deeply with theory and then we open PowerPoint or Illustrator and begin the meticulous process of turning our methods and findings into figures. We adjust arrows, realign boxes, standardize fonts, correct axis labels, and export multiple versions to meet submission guidelines. The science may be rigorous, but the path from text to publication-ready visuals often feels inefficient and cognitively draining. Lately, we’ve been looking at PaperBanana, a text-to-figure AI tool that’s clearly targeting researchers who want to speed up figure creation without sacrificing academic clarity. The core promise is simple: turn method descriptions into structured methodology diagrams, and turn data into charts in a way that’s meant to stay faithful to what we actually did (not just “look plausible”). The detail that makes it feel more “research-first” than generic image generators is the idea of a closed-loop workflow: instead of one-shot output, the tool aims to draft, check, and refine—so the figure is more likely to match the logic of the method. The “plus” here isn’t that we get a pretty figure in seconds. The real added value is that we get cheap iteration. When iteration is cheap, we stop freezing figures too early. We can generate multiple drafts, compare them, and treat figure-building like writing: draft → critique → revise. That’s where quality usually improves, not because the AI is perfect, but because our feedback loop gets faster and less exhausting. At the same time, it’s worth being honest: Illustrae plays a different game. It’s typically the better fit when the job is not just “create one diagram,” but assemble a full visual story, multi-panel figures, posters, teaching material, and lots of manual layout decisions. Illustrae tends to offer more features and flexibility, with more ready-to-use options to manage variables, layouts, iterations, and visual adjustments. The tradeoff many people feel is cost: it’s often described as significantly more expensive (and/or less predictable) than a straightforward researcher subscription, which can be a barrier for individuals or small labs. PaperBanana, in contrast, feels more minimalist and research-focused. For method diagrams and conceptual figures, especially at the drafting stage, it can be comparable in performance, while being more affordable, which matters a lot for labs, PhD students, and early-career researchers. It’s also commonly framed as being built on NanoBanana Pro, positioned as stronger for structured scientific visuals than general image models like DALL·E 3 (at least for diagram-like outputs where structure and labels matter more than “art style”). Here’s the comparison in a clean table (pricing stars are about budget-friendliness + predictability, not “how expensive the company is in absolute terms”): Tool Best for Pricing style (practically) Public price (USD, as listed) Price predictability (1–5) Value for researchers (1–5) Flexibility / control (1–5) Strengths (advantages) Tradeoffs (be honest) Illustrae Posters, multi-panel figures, teaching material, complex layouts Subscription + credits paid pricing not publicly posted (custom quote) (illustrae.co) 2 4 5 More features; more layout control; strong for assembling and polishing big visuals Can be significantly more expensive / less predictable; cost can block individuals/small labs; still needs expert review to avoid conceptual drift PaperBanana Methods diagrams, conceptual figures, fast drafting of research visuals Subscription + credits From $4.90/mo (annual billing) for 100 credits; $6.90/mo for 400; $19.90/mo for 1,500 (paperbanana.studio) (Alt. “credit plans” page also lists $14.90/mo, $59.90/mo, $119.90/mo tiers.) (paperbanana.org) 4 5 3 Minimalist and research-focused; fast drafting; good for method diagrams; affordable enough to be realistic for students/labs No meaningful free tier for deep exploration (subscription needed); final figures still need human refinement; confidentiality policies may not fully guarantee exclusion of unpublished work from training Now the limits, because this is where “AI for academia” either becomes useful or becomes risky. First, human review remains essential. These tools accelerate drafting, but final figures still need expert refinement: labels, implied causality, statistical meaning, and whether the diagram accidentally over-claims what the method can do. Second, subscription friction is real. If there’s no meaningful free tier, adoption becomes a budgeting decision, not a quick experiment, especially for students. Third, confidentiality is still a question. Unless a tool makes an explicit, strong guarantee that unpublished papers/figures are excluded from model training (and clarifies retention), we should be cautious with sensitive or pre-publication material. And yes, we can always rely on classic AI tools like Notebook or NotebookLM for summarizing, outlining, or restructuring ideas. They’re great at text workflows, but they’re not built specifically for researchers’ visual needs, and they’re typically less precise for scientific diagram conventions, which increases the risk of subtle visual or conceptual inaccuracies compared with tools designed for academic figures. So when we ask “what’s the plus?”, it’s this: we’re buying back attention. Not just time, attention. If figure drafting becomes fast enough that we can iterate without dread, we can redirect our effort to what actually moves research forward: clearer hypotheses, sharper methods, better interpretation, and figures that communicate rather than decorate. Choosing the right tool isn’t about hype, it’s about fitness for scientific purpose.

Si vous avez été en pratique assez longtemps, vous vous souvenez probablement comment vos premières années ont été façonnées par les parties peu glamour de l'emploi. Les notes, les rapports, les formulaires d'admission, les messages de planification, la notation et la documentation sans fin sont nettoyés. Il était épuisant, mais il faisait également partie de l'apprentissage. L'écriture des choses nous a forcés à clarifier ce que nous avons vu, ce que nous pensions que cela signifiait, et pourquoi nous avons choisi une étape particulière. Maintenant, l'intelligence artificielle entre dans cette couche exacte de la vie clinique à travers l'ergothérapie, l'orthophonie, la psychologie et la physiothérapie. Pas d'une façon dramatique « les robots remplacent les thérapeutes », mais d'une manière pratique et quotidienne. AI peut rédiger la documentation, résumer le texte long, organiser l'information et générer des modèles de premier passage. Cela change ce que les cliniques mesurent, ce que les gestionnaires attendent, et ce que les cliniciens de carrière se sentent pressés de livrer. De notre point de vue en tant que cliniciens, le plus grand changement est que l'IA compresse des parties de la courbe d'apprentissage. Un nouveau diplômé peut produire quelque chose qui semble poli très rapidement, et parfois que la production polie peut cacher le fait que le raisonnement clinique est encore en développement. Un paragraphe sûr n'est pas le même qu'une formulation solide. Un plan ordonné n'est pas le même qu'un plan individualisé. Lorsque la charge de travail est lourde et que le temps de supervision est limité, il devient plus facile d'erreur de vitesse pour la compétence, et c'est là que le risque augmente tranquillement. Donc la barre de niveau d'entrée est en mouvement. Si les tâches de routine deviennent plus rapides, l'attente devient que le clinicien contribuera davantage à ce qui ne peut être automatisé. Forte

Toward “Right‑Fit” Care: How AI Is Personalizing Mental Health Treatment Therapy has always included a careful kind of uncertainty, and as psychologists and mental health professionals we know how much skill it takes to work inside that uncertainty without trying to “solve” a person too quickly. We listen, assess, build a formulation, choose an evidence‑based starting point, and then we learn with the client what actually works. Even when we practice well, the early phase can feel like trial and error: a few weeks of testing whether this structure, this pacing, and this approach truly fit this person. For clients who are already exhausted or at risk, that delay matters. AI is beginning to shorten that “finding out” period by using personal data, shared only with explicit permission, to support earlier, more precise decisions. Instead of relying mainly on retrospective self‑report (“How was your week?”), we can add real‑world signals from phones and wearables: sleep duration and regularity, movement and sedentary time, daily rhythm, time spent at home versus outside, and changes in routine or social connection. In some specialized settings, clinicians and researchers also explore brain data (for example, MRI or EEG measures) to add information about brain circuitry and patterns that may relate to symptom profiles or treatment response. The aim is not to replace clinical judgment, but to strengthen it. The practical shift is moving from a snapshot at intake to a living picture of the client’s week. Self‑report is essential, but memory is imperfect and symptoms can blur recall. Passive and semi‑passive data can reveal patterns clients often feel but cannot easily name. If a client says they are “fine,” yet their sleep is fragmenting and their activity is steadily dropping, we have a compassionate entry point for deeper exploration. If anxiety spikes reliably at certain times and contexts, we can stop treating it as random and start treating it as predictable. This is where AI helps: it can analyze large, messy time‑series data and detect relationships humans would miss, what tends to happen before a mood drop, what predicts irritability, or what combination of isolation and sleep disruption precedes self‑harm urges. Think of it as a translation table from signals to clinical hypotheses. Sleep variability may indicate reduced emotion regulation capacity and relapse vulnerability. Reduced movement may point toward avoidance and anhedonia, suggesting behavioral activation or values‑based action. Abrupt routine changes may signal interpersonal rupture, shame, or safety concerns. The data does not diagnose; it helps us ask better questions sooner and refine the plan faster. You also pointed to a future‑leaning idea: combining brain scans with smartphone and wearable data to estimate the best intervention before a long course of trial and error begins. This direction is promising, but it demands caution. Some models can predict treatment response in research settings, yet they may not generalize across populations, devices, and real‑world complexity. Used ethically, these tools should function like decision support, a second opinion, never an automatic decision-maker. One of the most immediate benefits is timing. A growing class of tools aims to deliver support when symptoms are most likely to spike (often described as “just‑in‑time” interventions). Weekly therapy teaches skills, but the real test is whether clients can access them at 11 pm when exhausted, during a commute when panic builds, or right after conflict when urges rise. If data shows a reliable pattern, sleep disruption followed by next‑day agitation, or isolation followed by late‑night rumination, digital supports can be timed to the risk window: a brief grounding prompt, a coping-plan reminder, or a micro‑exercise that reconnects the moment to the formulation you built together. At their best, these tools feel like a bridge between sessions, not surveillance. These advancements could also expand access in a world of provider shortages. Not everyone can attend consistent sessions, and many people reach care only during crisis. Carefully designed digital supports can offer tailored continuity for those who struggle to access services, while keeping therapy human, relational, and collaborative when sessions occur. The ethical boundaries are non‑negotiable. Personal data must be opt‑in, purpose‑limited, and easy to pause or stop. The safest approach is minimalism: collect only what answers a clinical question. In many cases we do not need private content (messages, audio, contacts); we need patterns (sleep, movement, routine) and brief check‑ins. Whenever possible, the information should be anonymized or de identified (remove names, dates of birth, exact addresses, contact details, record numbers, and any unique identifiers) so it cannot reasonably be traced back to a person. The same logic applies to generative AI used for clinical writing or support: to protect confidentiality, it should ideally be local (downloaded and running on a device or secure internal server) rather than sending client information to an online system. If a cloud based tool is used, it should only be with explicit informed consent, clear limits on what data is entered, and a transparent explanation of where the information goes, who can access it, and how it is stored. And if a tool touches suicidality, the agreement must be explicit: what is monitored, who sees it, what triggers an escalation, and what the tool is not responsible for. Any system claiming it can “detect suicide risk” should be treated like a high‑stakes clinical claim requiring strong evidence, transparency, and a clear safety protocol. So how do we integrate this as therapists without losing the heart of therapy? Start small and clinical. Choose one target (sleep, panic spikes, avoidance, self‑harm urges). Collaboratively pick one or two metrics that feel helpful rather than invasive. Decide together how insights will be used, shaping session focus, planning for predictable windows, or evaluating whether a new intervention is working. Then review it like any intervention: Did it increase agency or self‑criticism? Did it clarify patterns or add pressure? Integration succeeds when the client feels more choice, more clarity, and faster relief. If we insist on evidence, close monitoring, clinical involvement, and regulation, AI can reduce unnecessary suffering by helping us match people

We keep noticing that when clinicians talk about “using AI,” we’re often talking about two very different approaches, even if we’re using the same tool. And the confusion usually shows up around one word: automation. People hear “automation” and imagine a cold replacement of therapy, or they assume it’s basically the same thing as collaboration. In practice, automation in clinical work is simpler and more grounded than that. It is not “AI doing therapy.” It is the clinician delegating repeatable steps in the workflow, then supervising the output the way we would supervise an assistant or a trainee. In procedural mode, AI becomes a substitute for execution. We ask, it answers; we paste, we send. The output is used for efficiency: quicker drafts, quicker wording, quicker structure. That can genuinely reduce load, especially on days when we’re holding multiple cases and still trying to document, plan, and communicate clearly. But procedural mode also has a built-in risk: it can bypass the step where we ask, “What claim did this just make, and do I actually have the clinical data to stand behind it?” In therapy, where work is high-stakes and context-sensitive, skipping that step is never a small thing. Collaborative mode looks different. Here, AI is treated more like a thinking partner that helps us refine what we already know. We provide context, constraints, and objectives, and we actively evaluate and revise what comes back. The benefit isn’t only speed, it’s quality. As goals become more complex, the work doesn’t disappear; it shifts upward into framing, supervision, and judgment. That shift is the point, because it mirrors what good therapy already is. The core value isn’t “doing tasks.” The core value is choosing what matters, staying accountable to the formulation, and tracking whether what we are doing is actually helping this client in this moment. With that clarity, the question “where does automation fit?” becomes easier: automation belongs around the session, not inside the relationship. It supports the repetitive work that quietly drains clinicians, so you show up with more focus and presence. In practical terms, this often starts with answering emails: drafting scheduling replies, boundaries, first-contact responses, follow-ups, and coordination messages with parents or schools. AI can give you a clean draft fast, but the clinician still protects tone, confidentiality, and the therapeutic frame before anything is sent. Automation can also support assessment workflows, especially the mechanical parts like cotation and report organization. It can help format tables, structure sections consistently, and draft neutral descriptions, saving time without pretending to “interpret.” Similarly, it can help with filling questions for you: generating intake questions, check-in prompts, or between-session reflection questions tailored to your model and the client’s goals. That doesn’t replace clinical judgment; it simply gives you a clearer scaffold for information-gathering and tracking change. Another high-impact area is session preparation. If you provide a brief, non-identifying summary of the last session, AI can help draft a focused plan: key themes to revisit, hypotheses to test, reminders of what was agreed on, and possible questions or interventions that match your orientation. The point is not to “script therapy,” but to reduce the mental load of reconstructing the thread so you can start the session grounded. More sensitive, but sometimes very helpful, is using automation around session recording and documentation (only with explicit consent and a privacy-safe system). AI can assist with transcripts, highlight themes, and draft a note structure or summary. Still, this must remain supervised: AI can miss nuance, misinterpret meaning, or phrase things too strongly. In clinical documentation, accuracy and accountability matter more than speed, so the clinician always verifies what’s written, especially around risk, safety planning, and any diagnostic or medical claim. Finally, automation can support what many clinicians want but struggle to do consistently: progress comparison over time. Whether you use outcome measures, session ratings, goals, homework follow-through, or narrative markers, AI can help summarize shifts from baseline, spot patterns across sessions, and draft a short “what’s improving / what’s stuck / what to adjust next” reflection. The tool organizes and surfaces patterns; you decide what it means and what the next clinical step is. All of this only works if we pay attention to data and privacy. We avoid entering identifying information unless we are using an approved, privacy-compliant system. We do not treat AI output as truth, especially for diagnosis, risk assessment, medication-related topics, or any medical claim. And we keep the clinician role explicit: AI can generate language, options, and structure, but we provide judgment, ethics, and accountability. This is also why many clinicians are drawn to running a private generative model locally on their laptop, offline, so data does not leave the device. Even then, strong device security and clear consent practices still matter, but the direction is sound: protect client information first, then build workflow support around it. When we use AI with this mindset, the payoff is real. We gain time and mental space for what cannot be automated: attunement, formulation, pacing, rupture-repair, and the relationship. The tool handles parts of the scaffolding and we protect the heart of therapy, which is slow, context-sensitive, and deeply human work.

If we’ve ever tried to write a real manuscript after a full day of research work, we already know the problem usually isn’t that we don’t have ideas. We do. The problem is that academic writing demands a specific kind of mental steadiness. We have to hold a thread, keep structure in mind, track definitions precisely, and build a coherent argument while our brains are already full of meetings, supervision, grant deadlines, data cleaning, reviewer comments, and the constant micro-decisions that come with running studies. So when OpenAI announced Prism, it caught our attention for a surprisingly practical reason. It sounds like it’s designed to reduce overwhelm in the writing process, not by writing the paper for us, but by making the environment less fragmenting and more supportive of sustained attention. OpenAI describes Prism as a free, cloud-based, LaTeX-native workspace for scientific writing and collaboration, with an AI assistant integrated into the document workflow. And that phrase, integrated into the workflow, matters. Many of us still write in a patchwork setup. The draft lives in one place, citations in another, PDFs in folders we swear are organized, tables in spreadsheets, figures in separate tools, and formatting rules that feel like a moving target. If we use AI, it often sits off to the side in a separate window, with no real awareness of what the document actually contains. Prism is pitching something different. One workspace where drafting, revising, compiling, and collaborating live together, so we don’t constantly switch contexts and lose momentum. That sounds less like automation and more like good research infrastructure. Something that helps us keep the argument intact while we spend our limited energy on what actually matters: methods, logic, interpretation, and the discipline of not overclaiming. What we also appreciate is that Prism seems aimed at the boring practical problems that quietly wreck productivity. Collaboration, comments, proofreading support, citation help, and literature workflow features are not flashy, but they are exactly the kind of friction that makes us close the laptop and tell ourselves we’ll do it tomorrow because we can feel the administrative drag consuming what’s left of our focus. And if we’ve ever co-authored a paper, we know how much time gets lost to version control, merging edits, and re-checking what the “current draft” even is. A shared cloud workspace can reduce that overhead by keeping writing and collaboration in one place. Here is where the researcher angle comes in. Researchers are trained to track nuance, uncertainty, and the limits of what data can actually support. Many of us can write well when we have space to think. But research rarely gives writing prime attention. Writing happens in stolen hours between analyses, teaching, project management, and funding applications. That changes what “helpful technology” looks like. We don’t just need a tool that generates text. We need a tool that helps us stay oriented so we can turn results into clear contributions publishable, teachable, and useful. Prism might support that kind of work, especially for researchers who publish, teach, supervise trainees, or collaborate across institutions and need their writing process to be less chaotic. If it truly reduces friction, it could help more of us finish what we start—not because the tool has better ideas than we do, but because it helps protect the continuity of our thinking. At the same time, we should say the quiet part out loud. A smoother writing workflow doesn’t automatically mean better science. AI can help us sound coherent and academic, and that can be useful, but it is also where risk shows up because polished writing can hide weak reasoning. So if we use Prism, we should treat it like a very fast assistant. It can reduce friction and help us express what we mean, but it is not the source of truth. We still own the reasoning, the claims, the citations, and the integrity of the work. And of course, Prism is not the only tool that exists. Most of us have already used other AI tools before, along with specific writing and reference managers that keep our workflow moving. What makes Prism feel different, at least from the way it is described, is the promise of one integrated workspace and the fact that it is free. If it delivers even half of that, we honestly cannot wait to explore it more. Where we land is simple. Prism sounds promising because it aims at the real pain points in research writing: context switching, formatting drudgery, collaboration friction, and the cognitive load of keeping a complex document coherent over time. Not magic. Not a replacement for expertise. But possibly the most researcher-friendly kind of productivity tool—the kind that helps us keep the thread.