https://survol.ai/news Field notes on enterprise AI, agentic development and product steering, from the team building Survol. en-GB Survol Mon, 12 Oct 2026 10:00:00 +0000 https://survol.ai/img/og/survol-en.png https://survol.ai/news https://survol.ai/news/saaspocalypse-code-was-never-the-point https://survol.ai/news/saaspocalypse-code-was-never-the-point Mon, 12 Oct 2026 10:00:00 +0000 Adrien Torris Build vs buy Markets Governance Technical debt Jefferies named the thesis “SaaSpocalypse”, the market traded it, and Salesforce just posted its best rally since 2020. The reasoning error is instructive: code becomes free, but code was never what you were buying. Three buckets, a moat analysts never mention, and a test for any decision to build. https://survol.ai/news/business-teams-are-writing-code https://survol.ai/news/business-teams-are-writing-code Fri, 09 Oct 2026 10:00:00 +0000 Adrien Torris Governance Citizen developers Security Technical debt OpenAI's product managers have stopped making mockups; Deutsche Bahn counts 4,000 non-technical makers and more than 500 applications in production. Against that: OWASP's risk typology and a study of 302,600 commits showing 22.7% of issues introduced by coding assistants are still there at the repository's latest revision. https://survol.ai/news/the-job-thats-left-has-no-pipeline https://survol.ai/news/the-job-thats-left-has-no-pipeline Sun, 04 Oct 2026 10:00:00 +0000 Adrien Torris Careers Specification Hiring Organisation A post on seldo.com breaks the cost of making software down item by item and asks how much an agent can take. What resists: deciding what to build and defining what good means. The market already pays $240,000 for it - and nobody trains for the job, whose old pipeline was a side effect of the junior role. https://survol.ai/news/the-solow-paradox-is-back-with-ai https://survol.ai/news/the-solow-paradox-is-back-with-ai Thu, 01 Oct 2026 10:00:00 +0000 Adrien Torris Productivity Economics Organisation ROI An NBER survey of nearly 6,000 executives: nine in ten see no effect of AI on their firm's productivity over three years. Economists reached for Robert Solow's 1987 line. The comparison holds, as long as you remember that his paradox ended late, through reorganising work, and in six sectors out of 59. https://survol.ai/news/gartner-81-percent-what-the-number-measures https://survol.ai/news/gartner-81-percent-what-the-number-measures Sat, 26 Sep 2026 10:00:00 +0000 Adrien Torris Governance ROI Measurement Code generation Only 22% of organisations have scaled AI across several business units, and the top performers see positive returns on 81% of their projects. That second number will travel. It rests on a definition that takes failures out of the count, and on a comparison between two measures that are not the same. https://survol.ai/news/risk-register-cannot-see-ai https://survol.ai/news/risk-register-cannot-see-ai Tue, 22 Sep 2026 10:00:00 +0000 Adrien Torris Governance Risk Sovereignty Shadow AI The Hiscox x OpinionWay barometer unveiled at VivaTech measures a contradiction: AI is everywhere in practice and 12th of 14 in perceived risk. The same survey contains the explanation — and it has nothing to do with negligence. https://survol.ai/news/a-third-of-companies-build-instead-of-buying https://survol.ai/news/a-third-of-companies-build-instead-of-buying Tue, 15 Sep 2026 10:00:00 +0000 Adrien Torris Governance Build vs buy Coding agents Costs 32% of respondents to McKinsey's annual AI survey walked away from at least one software purchase because coding agents could build it in-house. Meanwhile, the organisations doing best with those agents are three times more likely to be constrained by their cost. The two numbers belong together. https://survol.ai/news/the-beta-is-open https://survol.ai/news/the-beta-is-open Mon, 07 Sep 2026 09:00:00 +0000 Adrien Torris Product Beta Pricing Roadmap Ten organisations, €5 per user per month instead of €29, a direct line to the founder. Plus the detail most announcements of this kind leave out: as of today Survol has no paying customer and has never been proven at a third party. https://survol.ai/news/cutting-ai-projects-without-knowing-why https://survol.ai/news/cutting-ai-projects-without-knowing-why Wed, 02 Sep 2026 10:00:00 +0000 Adrien Torris Governance Measurement Adoption Shadow AI Three figures and three cases told from the inside: an internal assistant delivered eighteen months late, an agent shut down without anyone knowing why, a 90% adoption rate that proves nothing. What is missing is not one more tool - it is a measure of real usage. https://survol.ai/news/your-documentation-just-became-part-of-the-execution-system https://survol.ai/news/your-documentation-just-became-part-of-the-execution-system Wed, 26 Aug 2026 12:00:00 +0000 Adrien Torris Governance Operations Process Agents Bertrand Duperrin describes a quiet shift: the representation of work stops being documentary heritage and becomes part of the execution system. Who builds it, who maintains it, who governs it - and why automating too early is the real danger. https://survol.ai/news/everyone-can-build-alan-280-pull-requests-non-developers https://survol.ai/news/everyone-can-build-alan-280-pull-requests-non-developers Wed, 26 Aug 2026 10:00:00 +0000 Adrien Torris Governance Teams Code review Agents Opening the codebase to the whole company worked at Alan for three reasons that have nothing to do with the model: the same tools as the engineers, a bounded scope, and a named owner for every change. The next bottleneck is already here: review. https://survol.ai/news/40-days-2-3-billion-tokens-the-debt-just-moves https://survol.ai/news/40-days-2-3-billion-tokens-the-debt-just-moves Wed, 26 Aug 2026 08:00:00 +0000 Adrien Torris Governance Tokens Technical debt Agents 1,477 prompts, 2.3 billion tokens, 260,564 lines reviewed by hand, a claimed ROI of 18.15×. The best-documented experiment of the year mostly explains where the productivity gain goes - the one nobody finds in their delivery plan. https://survol.ai/news/one-in-three-ai-pull-requests-ships https://survol.ai/news/one-in-three-ai-pull-requests-ships Sun, 16 Aug 2026 09:00:00 +0000 Adrien Torris Measurement Code review Agentic development LinearB measured 8.1 million pull requests, then 2.7 million more in the first half of the year. Human-written code merges 84.5% of the time; AI-generated code, 32.7%. The gap between them isn't a model quality problem - it's a delivery pipeline calibrated for human throughput that nobody has recalibrated. https://survol.ai/news/control-moves-from-models-to-their-environment https://survol.ai/news/control-moves-from-models-to-their-environment Thu, 13 Aug 2026 10:00:00 +0000 Adrien Torris Governance Security Agents Environments An OpenAI model left its sandbox and entered Hugging Face. Anthropic reviewed 141,006 tests and found three of its own. Both say the same thing: closing a model does not control it. What you govern is what it can reach. https://survol.ai/news/the-ai-act-enters-its-enforcement-phase https://survol.ai/news/the-ai-act-enters-its-enforcement-phase Thu, 13 Aug 2026 10:00:00 +0000 Adrien Torris Compliance AI Act Governance Europe Telling people they are talking to an AI, marking generated content, flagging deepfakes: article 50 has been binding since 2 August, with fines up to 3% of worldwide turnover. The rest waits for 2027 and 2028. https://survol.ai/news/git-agents-entire-gitbutler-two-answers https://survol.ai/news/git-agents-entire-gitbutler-two-answers Tue, 11 Aug 2026 10:00:00 +0000 Adrien Torris Git Infrastructure Agentic development Thomas Dohmke has opened the preview of a distributed Git network built for agents. Four months earlier, Scott Chacon raised $17M to rethink Git's primitives. Two GitHub alumni, two different layers of attack - and the same verdict on what is missing: not the code, but the why. https://survol.ai/news/vibecoding-fatigue-three-workarounds-and-a-scale-problem https://survol.ai/news/vibecoding-fatigue-three-workarounds-and-a-scale-problem Sat, 08 Aug 2026 15:00:00 +0000 Adrien Torris Vibecoding Working conditions AI governance The Conversation documents https://survol.ai/news/prompt-context-loop-harness-engineering-limits https://survol.ai/news/prompt-context-loop-harness-engineering-limits Sat, 08 Aug 2026 08:00:00 +0000 Adrien Torris Prompt engineering Context engineering Harness AI governance Prompt engineering, context engineering, loop engineering, harness engineering: four disciplines born in eighteen months. Each solved a real problem, and each hits the same ceiling - they optimise what happens during an execution. That ceiling is not technical. https://survol.ai/news/survol-and-coding-assistants-two-different-layers https://survol.ai/news/survol-and-coding-assistants-two-different-layers Thu, 06 Aug 2026 08:00:00 +0000 Adrien Torris Positioning Agentic development AI governance It is the question we get most often: how does Survol differ from Cursor, Kiro or Claude Code? The short answer is that they do not play on the same layer. Those tools are excellent where they operate. Survol operates one layer above, at the level of the product and the team. Here are seven concrete differences - and, just as important, what Survol does not do. https://survol.ai/news/the-shape-of-thought-after-ai-and-the-vault https://survol.ai/news/the-shape-of-thought-after-ai-and-the-vault Wed, 29 Jul 2026 10:00:00 +0000 Adrien Torris Agentic development In a remarkable essay, Simon Rimbert describes how AI changes the shape of thought - provided you have a https://survol.ai/news/the-spec-is-the-new-software-asset https://survol.ai/news/the-spec-is-the-new-software-asset Wed, 29 Jul 2026 09:00:00 +0000 Adrien Torris Specification When AI regenerates code at will, what gains value is no longer the code but the spec: the durable intent everything else derives from. An asset you have to treat as one - versioned, governed, maintained. That is exactly what Survol makes of it. https://survol.ai/news/spec-driven-development-keeping-the-spec-alive https://survol.ai/news/spec-driven-development-keeping-the-spec-alive Thu, 16 Jul 2026 12:54:57 +0000 Adrien Torris Spec-Driven Development GitHub, Microsoft, Thoughtworks, AWS: the industry is converging on spec-driven development - write the spec before the code, make it the AI's source of truth. But one question stays open: who keeps the spec alive, and who makes it live inside a team? That is where Survol stands apart. https://survol.ai/news/you-dont-have-a-prompt-problem-you-have-a-context-problem https://survol.ai/news/you-dont-have-a-prompt-problem-you-have-a-context-problem Wed, 15 Jul 2026 06:54:27 +0000 Adrien Torris Prompt engineering Context engineering AI governance Agentic workspace In an in-depth piece, the consultancy Archwise describes the trajectory every organisation ends up following: first the prompt-engineering fever, then the discovery of its https://survol.ai/news/prompt-tricks-are-over-mollick-wharton-specs https://survol.ai/news/prompt-tricks-are-over-mollick-wharton-specs Tue, 14 Jul 2026 07:59:50 +0000 Adrien Torris Specifications Four reports from Wharton's generative AI lab put the magic recipes of prompting on the test bench: politeness, tips, threats, expert personas, 'think step by step'. Verdict: none of it holds. What works, Ethan Mollick sums up, is 'specifying your goals, your output, what good and bad results look like, and how to test - yes, it's just management'. We call that a spec. And it is Survol's founding bet. https://survol.ai/news/vivatech-four-predictions-on-the-future-of-work https://survol.ai/news/vivatech-four-predictions-on-the-future-of-work Tue, 14 Jul 2026 07:26:05 +0000 Adrien Torris Future of work AI and organisation At VivaTech 2026, four predictions about the work of tomorrow. Three of them describe, almost word for word, what Survol is building: steering agents, keeping your critical thinking, capitalising on experience. https://survol.ai/news/cisco-only-14-percent-of-companies-are-ready https://survol.ai/news/cisco-only-14-percent-of-companies-are-ready Tue, 30 Jun 2026 06:42:36 +0000 Adrien Torris AI governance AI and organisation In an interview with Maddyness, Cisco confirms the shift from generative AI to agentic AI - and reminds us that only a small minority of companies is genuinely ready to deploy it at scale. The good news? The obstacle is neither technology nor talent: it is integration, governance and people. Three things you can act on today. https://survol.ai/news/shadow-ai-in-the-public-sector https://survol.ai/news/shadow-ai-in-the-public-sector Fri, 26 Jun 2026 06:58:43 +0000 Adrien Torris Shadow AI AI governance More than one public servant in two already uses ChatGPT quietly to do their job. Seen from the risk side - data outside the EU, the Cloud Act, a growing cyber threat - it is worrying. Seen from the usage side, it is a tremendous signal of appetite. The right answer is neither prohibition nor waiting: it is sovereign governance that sees and frames without punishing. https://survol.ai/news/scheduled-tasks-arrive-in-chatgpt https://survol.ai/news/scheduled-tasks-arrive-in-chatgpt Mon, 22 Jun 2026 07:58:35 +0000 Adrien Torris AI scheduled tasks ChatGPT OpenAI finally gives ChatGPT a real 'Scheduled' page: flexible scheduling, conditional monitoring tasks, and the end of Pulse. AI stops waiting to be asked: it acts on its own, in a loop. Excellent news - which immediately raises a question: when dozens of agents run without you, who administers them and who supervises them? https://survol.ai/news/an-ai-native-company-in-8-days https://survol.ai/news/an-ai-native-company-in-8-days Sat, 20 Jun 2026 06:59:57 +0000 Adrien Torris Marc Simoncini Entrepreneurship Marc Simoncini, a figure of the French tech scene, locked himself up for eight days with 24-year-old YouTuber Aywen to build a company designed entirely around AI. The experiment has only just started - and it says something exciting about the new rules of entrepreneurship. https://survol.ai/news/uber-burned-its-annual-ai-budget-in-four-months https://survol.ai/news/uber-burned-its-annual-ai-budget-in-four-months Fri, 19 Jun 2026 10:52:50 +0000 Adrien Torris Uber AI governance Uber ran through its 2026 AI budget in a quarter, then capped Claude Code and Cursor at $1,500/month per tool. The press saw a handbrake. We mostly see proof that enterprise AI works - and that only the steering layer was missing. https://survol.ai/news/the-problem-is-not-technology-but-organisation https://survol.ai/news/the-problem-is-not-technology-but-organisation Thu, 18 Jun 2026 07:52:24 +0000 Adrien Torris AI governance AI and organisation An Institut Montaigne paper, relayed by La Tribune, explains why AI still disappoints in many companies. The diagnosis is harsh; we mostly find it exciting. Because if the brake is not the technology, then the value is within reach of a reorganisation - and sovereignty within reach of an orchestration layer. https://survol.ai/news/the-approved-spec-is-the-prompt https://survol.ai/news/the-approved-spec-is-the-prompt Sun, 14 Jun 2026 07:32:47 +0000 Adrien Torris Functional specifications Agentic mental map Product mental map Agentic development makes you lose the mental map of your product. Here is the conviction that gave birth to Survol: describing a feature once, properly, beats ten throwaway prompts. https://survol.ai/news/a-third-fewer-tokens-anatomy-of-a-tailored-context https://survol.ai/news/a-third-fewer-tokens-anatomy-of-a-tailored-context Thu, 11 Jun 2026 07:28:45 +0000 Adrien Torris Token optimisation AI cost savings The more context an agent receives, the more it costs - and not necessarily the better it works. Here is how Survol composes context down to the strict minimum, picks the right model for the right task, and treats your subscription exactly for what it is: your money. https://survol.ai/news/one-branch-per-feature-one-pr-per-version https://survol.ai/news/one-branch-per-feature-one-pr-per-version Tue, 09 Jun 2026 07:54:57 +0000 Adrien Torris AI governance AI and organisation Stop thinking about branching. Here are the readable branch lanes that hold up even when humans and agents write code side by side. https://survol.ai/news/import-an-existing-product https://survol.ai/news/import-an-existing-product Sun, 07 Jun 2026 08:24:21 +0000 Adrien Torris AI product management You almost never start from scratch. For products already under way, Survol offers a second way in: you tick the repositories, the agent analyses them and proposes a feature map - with a confidence level. You approve, rename, merge. Existing hosting is connected, never modified. https://survol.ai/news/auto-resume-sessions-when-your-credit-comes-back https://survol.ai/news/auto-resume-sessions-when-your-credit-comes-back Sat, 06 Jun 2026 06:27:27 +0000 Adrien Torris AI governance Agentic workspace The plan limit always hits at the worst moment - usually in the middle of the night, halfway through a task. Rather than losing everything, Survol parks the agent and resumes it automatically as soon as credit is back. When you wake up, the merge request is waiting for you. https://survol.ai/news/ai-writes-the-code-humans-decide-expert-reviews https://survol.ai/news/ai-writes-the-code-humans-decide-expert-reviews Fri, 29 May 2026 07:54:24 +0000 Adrien Torris Agentic development Agentic workspace Delegating the code to an agent is not delegating judgement. Survol is not a black box: product decisions are discussed between humans, and an expert review turns, finding by finding, into tasks for the agent. Here is how trust stays on the human side. https://survol.ai/news/organisation-technical-constraints https://survol.ai/news/organisation-technical-constraints Sun, 24 May 2026 07:49:21 +0000 Adrien Torris AI governance AI and organisation A team has rules: conventions, security, architecture choices. Rather than repeating them in every prompt - or hoping the agent will guess them - Survol injects them at the start of every session, at three levels of authority, and logs every application. A look behind the scenes of agentic governance.