An agent, given its full history, finished fewer tasks than one given no memory. That's the striking result on Apple's research page today. Read the current paper though, and the authors make a narrower claim. Compact context costs less, while the completion evidence is still small. Before your team decides to remember everything or delete the lot, test which information the next job actually needs. This is the Augur Dispatch for Thursday, September 17, 2026. The Enterprise study covered 24 tasks. Selective memory completed 23, no memory, 19, and full history, 17. The authors say those differences aren't statistically established after their correction for multiple comparisons. There's a second complication. In 18 of the selective condition tasks, the system reused an existing artifact and updated its data without calling the model. That's useful automation, but it makes this more than a comparison of memory settings. The sensible response is a bounded pilot, not a company-wide order to forget more. Picture the analyst who rebuilds a weekly sales report because the agent has forgotten which columns matter and how the director wants the result laid out. The useful memory might fit on a page, the task, the data structure, the tools available, and the rules for the output. That's what the design retains. The conversation's working notes and reasoning traces don't come along. In the paper's public data experiment, Compact Memory completed all 12 trials, full history, 8, and no memory, none. But the last group wasn't given the earlier formatting requirement again. Its reports worked structurally. They failed a requirement the agent couldn't see. Forgetting the instructions is a fairly dependable way to miss the instructions. The difference between compact memory and full history wasn't statistically established on completion either. Compact memory used about half the input tokens, which are the pieces of text the model processes and providers commonly charge for. The compact workspace also retained the earlier artifact, so this experiment doesn't cleanly separate the value of memory text from the value of having existing work to revise. These are the author's experiments, not an independent replication. That still gives a workflow in or something concrete to investigate. A transcript may carry useful context, but it may charge you repeatedly for a lot of material the next task doesn't need. Saving less also means deciding who gets to save anything. A memory injection paper describes how one poisoned interaction can steer an agent's later answers without the attacker directly editing its memory store. It doesn't show that everyone sharing a workspace becomes compromised. That's a separate test, not a finding we can borrow. Imagine a team sharing a reporting workspace. A convenient instruction saved on Tuesday could become Thursday's unwanted rule, unless someone owns the review and knows how to reverse it. The Apple hosted design includes permissions tied to people's roles. The practical question is whether those permissions cover the right people and the right changes. A short memory can still contain a bad instruction. The commercial products aren't simply offering a bigger transcript bucket either. Anthropic describes individual memory topics and separate project memory. Sakana lets people review retained preferences and turn memory off. Their launches make continuity easier to use. They don't prove that either vendor has abandoned selection. IBM adds another reason to avoid a universal rule. Its experiments concern reusable guidelines distilled from earlier work, not whole transcripts. Some stronger models benefited from the full guideline set, while other models did better with selective retrieval. The amount worth carrying depends on the model and the job. Buying everyone a larger filing cabinet won't settle that. For a useful first test, choose one recurring report and put its workflow owner in charge. Have a security reviewer approve what can enter shared memory and review the results on September 24th. Compare the existing workflow with no memory, the full transcript, and a compact page of instructions, keeping the model, tools, tasks, and scoring consistent across the comparison. Give each version the information needed for the current task. If a requirement must carry over from an earlier session, make that distinction explicit rather than quietly denying one version the instructions. Count correctly completed reports, then measure model usage and the time a person spends fixing them. Test refreshing an existing report separately from generating a new one. Keep the compact version only if it maintains quality while lowering cost or correction time. Stop for a material quality drop, an unauthorized memory change, or exposure of sensitive data. Expand only after the result repeats on held out cases and another workflow. If the transcript wins, find out what useful information your page left out. You haven't proved that the agent needed its earlier reasoning merely because that reasoning was in the winning transcript. Any lasting advantage comes from knowing your workflow well enough to write and maintain those instructions, and from having realistic tests. Your competitors can buy the same memory feature. Control over context leads to a different purchasing question: control over the supplier. Cohere and Aleph Alpha have signed an agreement to combine under the Cohere name with planned headquarters in Berlin and Toronto. Regulatory approval is still required, and the companies expect closing later this year. A procurement lead should distinguish the supplier available to sign today from the combined company described in the announcement. Ask Council how a change in ownership affects the agreement. Mistral's announcement with Mozilla concerns a product already in use, Firefox Smart Window, still in beta. It says Mistral models serve users in France and North America, with the United Kingdom and Germany expected later this year. Partners such as Mistral agree to zero data retention for those interactions. That is a specific commitment for this service, not a blanket promise about everything carrying the vendor's name. Compliance teams also need to distinguish political statements from enforceable rules. The dispatch's cited commentary reports President Trump calling extinction warnings a hoax. That doesn't settle what Washington will do next. The European AI office has enforcement powers over general-purpose models from August 2nd. Build around the obligations that apply to you while treating the American policy direction as unsettled rather than absent. A lower model bill means its own evidence. Together AI says some customers cut costs by up to 70% when moving to open models and suggests starting with 10% of traffic. That's a vendor-reported result and a migration suggestion, not a saving you can put into next quarter's budget without checking your workload. Watch the costs per correctly completed task, including whatever people have to repair. Sakana has made Fugu Max available to all Sakana chat users, replacing the earlier Fugu option. Users select it from the model picker. It routes requests across open models. Frontier level performance remains Sakana's claim. On Augur's own record, the August call that Together, Fireworks, or OpenRouter would host DeepSeek V4 Flash by September 15th is graded a provisional miss. That stays on the record. Today's new calls are testable predictions with explicit deadlines, not recommendations to buy a product. Augur assigns 60% odds that Cohere publicly announces the completed Aliph Alpha combination by December 31st, 2026. A public closing confirmation from either company qualifies. No confirmation by the deadline or termination is amiss. The other call carries 55% odds. Mozilla or Mistral publicly confirms Smart Window with Mistral models is available in both the United Kingdom and Germany by that same date. Both countries must be live. Either still unsupported or delayed beyond the deadline means amiss. For now, pilot compact memory on one workflow. Watch the deal closing and the regional rollout. Don't turn a small research result into a deletion policy. Which recurring task would you trust to a page of maintained instructions? And what would you need to see before dropping its transcript? That is the Augur Dispatch for Thursday, September 17th, 2026. Thank you for listening. The written brief with every source is at augurdispatch.ai