
人間はAIエージェントとどう協働するか — Atlassian CEO マイク・キャノン=ブルックス
How Humans Will Work Alongside AI Agents — With Mike Cannon-Brookes

人間はAIエージェントとどう協働するか — Atlassian CEO マイク・キャノン=ブルックス
Big Technology Podcast
要約
Atlassian共同創業者兼CEOのマイク・キャノン=ブルックス氏が、Web Summitの公開収録でAIによる仕事の変化を語った回。雇用は消えるより「変わる」との見方を示し、人とエージェントの協働形態や、Teamwork Graphによる文脈提供、モデルのコモディティ化論、ソフトウェアの「ヘッドレス化」への批判を述べている。
この回のポイント
- ●キャノン=ブルックス氏は、AIで職が消えるかより、仕事の中身がどう変わるかが重要だと述べ、銀行支店の例で説明した。
- ●AIによる効率改善はまだ期待ほどではなく、モデルの進歩が成果に結びつかない理由は主にデザイン・体験と人の考え方の変化にあるとの見方を示した。
- ●人とエージェントの関わり方を、チャット、委任(delegation)、共同編集的なマルチプレイヤー、隣り合って作業するペアリングに分けて整理した。
- ●Teamwork Graphは約8年かけて構築した企業向けコンテキストグラフで、エージェントのセッションから学習内容を抽出して次回に活かし、トークン消費の削減にもつながるという。
- ●モデルは完全には汎用品化せず、特性やコストに差が残るため、モデルの使い分け(ブレンディング)とハーネスが重要だと語った。
- ●ヘッドレス(UIを捨ててエージェント経由のみにする)戦略は「脳がない」選択で、UIへの投資を続けるべきだと主張した。
章立て
仕事の変化と雇用への見方
エージェント導入で効率・品質・創造性がどう変わっているかと、レイオフはAIというより再構築だという見解を語る。
銀行支店の例と職種の変化
銀行支店の事例を挙げ、職は消えるより中身が変わると説明。職種間の比率が変わる可能性にも触れた。
生産性が伸びない理由とAMP
成果が出にくい理由は人の考え方とUI設計にあるとし、AtlassianのAgentic Multiplayer Protocol(AMP)を紹介した。
未来の仕事と協働の形
タスクより成果を考える働き方、抽象度の引き上げ、チャット・委任・協働・ペアリングという四つの形態を解説。
Teamwork Graphの役割
企業の知識や成果物を束ねるグラフを、人間の検索とエージェントのMCP/CLIの両方に提供し、学習で改善していく仕組みを説明。
モデルのコモディティ化と競合
モデルは差が残るが重要度は過大評価されているとし、OpenAIやAnthropicなどモデル提供企業の参入への見方を述べた。
ヘッドレスは脳なし
ソフトウェアのヘッドレス化を批判し、MCPへの投資とUI設計の継続を両立させるべきだと主張した。
解説記事
Atlassian(Jira、Confluence、Loomなどを提供)の共同創業者兼CEO、マイク・キャノン=ブルックス氏が、Big Technology PodcastのWeb Summit公開収録に登場した。AIが仕事とそれを支えるソフトウェアをどう変えるのかについて、雇用、協働の形、競争環境、UI戦略の観点から語っている。
雇用は「消える」より「変わる」
冒頭で司会者は、AIが仕事を奪うならすでに表れているはずだが、レイオフした企業は「AI時代に向けた再構築」と説明する点を指摘した。キャノン=ブルックス氏は、再構築という説明のほうがはるかに正確だと考えていると述べる。5年後を断言する人は信用すべきでないとしたうえで、差し引きでは職は変わる数よりも多く生まれるという見方を示した。
例えとして挙げたのが銀行支店である。2000年代にはネット・モバイルバンキングで支店は消えると言われたが、過去10年は豪州や米国で支店が増えており、しかも2000年当時とは役割がまるで違うという。ソフトウェア開発者もデザイナーもマーケターも残るが、やり方は大きく変わる。プロダクトマネージャーが増えるのかエンジニアが増えるのか、職種間の比率は「やってみないと分からない」とも語った。
生産性が伸びにくい理由
同氏は自社を「AI製品を売る会社」ではなく、ワークフローを回す技術製品の会社だと位置づける。AIはモバイルやインターネットと同様に新しい技術要素だという考えだ。
モデルが進歩しているのに企業の成果へ十分反映されない理由について、答えの多くはデザインと体験にあると述べる。人が仕事の考え方を変えることと、製品のインターフェースが大きく変わることが両輪で必要だという。同社が約1か月前にアムステルダムのカンファレンスで発表したAMP(agentic multiplayer protocol)は、エージェントをチームに迎えるための体験・技術・パターンの集合で、内部ではMCPやA2Aなどの標準を使っていると説明した。
仕事の中身については、作業の手順より「得たい成果」を考える比重が増えると述べる。たとえば財務モデルなら、AIが作った結果が正しいかどうかをどう検証するかを考える必要がある。また、質問して答えを得る形から、ループを回して成果に近づけていく、より高い抽象度の使い方へ移るという。ただし誰もがそうした専門的なエンジニアになる必要はなく、そのための考え方をアプリのUIに組み込むことが課題だとした。
人とエージェントの四つの関わり方
同氏は協働の形態を整理している。一つはチャットのような一対一のやり取り。二つ目は委任で、会議に自分の代理としてエージェントを送り出し、あとで結果を聞く形である。三つ目は、Confluenceのページを複数の人や複数ベンダーのエージェントが同時に編集するマルチプレイヤー型。四つ目はペアリングで、会議机の同じ側にエージェントと並んで座り、同じ成果に向けて横並びで作業するイメージだ。
どの形態が適切かは製品によって異なり、Loomではエージェントに動画を作らせたり、届いた動画を見てコメントを返したりする場面もある。全員がコマンドラインでやり取りする未来は想定していないと述べた。
Teamwork Graphとモデルの位置づけ
AMPの下の層にあるのがTeamwork Graphで、約8年前に5人ほどで始まり、現在は約1000人が関わる企業向けのコンテキストグラフだという。文書、プロジェクト、メール、会議の文字起こし、ソースコード、データ資産などを一つのグラフに束ね、人には検索やチャットで、エージェントにはCLIやMCP経由で答えを返す。エージェントがこれを使うと回答の質が上がり、トークン使用量も減って速くなると説明した。さらにエージェントのセッションから重要な記憶や学びを抽出し、次回の検索結果に反映することで、改善が積み上がるとしている。
司会者がモデルの話が出なかった点を指摘し、モデルはコモディティ化していると問うと、同氏は完全にそうは思わないと答えた。モデルごとに得意分野やコスト特性が異なり続けるが、重要度は世間で言われるほどではないという。最新モデルに飛びつくのは高価で遅いことも多く、どのモデルがどの問いに強いかを見極める「モデルブレンディング」が重要で、その上にハーネス、コンテキスト、アプリケーションの層が重なると述べた。
モデル企業が上流に進出し、組織のソフトウェアを競合として作る懸念については、「心配という言葉は適切でない」と応じた。かつてMicrosoft、Google、Amazonについても同じ議論があったと振り返り、集中力や専門性という制約はどの企業にもあるとの立場だ。また、AIでコードのコストが下がったのは自社も同じで、モデルの進歩は各ベンダーにほぼ平等に行き渡るとも述べている。
「ヘッドレスは脳なし」
最近はエージェント経由でのみソフトウェアを使わせる「ヘッドレス」化が語られる。同氏はこれを「brainless(脳なし)」と呼び、UIの設計・体験への投資をやめるのは衰退への道だと批判した。AtlassianはCLIやMCPサーバーにも大きく投資しており、そのMCPサーバーは主要ラボで最も使われているものの一つだとしつつ、それとUI投資は両立すると説明する。
理由として、多くの人が企業向けソフトをターミナルだけで使うのは非効率で、ウィンドウやメニューのあるGUIが生まれたのには理由があると述べた。先週出た調査でも、そうした働き方は生産的でないと人々が感じていたという。生成AIによる動的UIには可能性を見ており、社内で試作もしているが、画面が毎回まったく変わると使いにくいとも指摘した。
まとめ
今回の議論の要点は、AI活用の成否が技術だけでなく、人の考え方とインターフェース設計に左右されるという点にある。日本の企業にとっても、人員削減の是非より、職務の中身をどう組み替えるかが問われるという示唆がある。また、MCPなどでエージェント接続を整えつつUI体験にも投資する、というSaaSベンダーの戦略論は、自社製品の設計を考える開発者や事業責任者にとって参考になるだろう。
文字起こし(英語・自動生成)
How do you price intelligence as a service? How do you charge for delivering end-to-end work, not just access? Those questions remain front and center for AI founders and operators, and getting to the right fit business model, whether that's selling usage credits, outcomes, seats, or a hybrid combination, requires constant iteration. Chargebee is the billing and monetization infrastructure that powers just that for ambitious, fast-scaling AI businesses. Visit chargebee.com slash Alex to learn more. That's C-H-A-R-G-E-B-E-E dot com slash Alex. I'm Alex Cantoros. I'm the host of Big Technology Podcast, and I'm here with Mike Cannon-Brooks, the CEO of Atlassian, for a conversation about how AI changes the world of works. And we're recording live at Web Summit with a large audience. Now, folks, this is going to go on the podcast feed and on YouTube, and I want to make sure everybody has a good idea of how many people are watching. So please, make some noise.
I said make some noise, especially these folks in the back. Let's go. Mike, it's good to see you. Nice. Thanks for having me. We're talking at a very interesting time. AI, in terms of its ability to change work, was always talked about as something theoretical. Something that may happen in the future, that we'll see the impact of it, and folks basically speculated. But today, AI is in the hands, not just of thousands or hundreds of thousands, but of millions of people. And as the CEO of Atlassian, with brands like Jira and Trello, probably no one has a better look than you into the way that work is changing. So can you just give us your perspective on, based off of what you're seeing, how the nature of work is changing today? Work is changing. Look, the obvious answer is we're adding a lot of agentic work into our workflows. If you think about any business process, we are finding small and bigger ways to put AI, agents, harnesses,
whichever way you want to refer to them, into those processes. We're getting some efficiency gains, arguably not as much as people would like to see in a lot of workflows yet. We are seeing some quality improvements, so the output of those processes is improving. And I think we're seeing the start of a big growth in creativity, so the amount of things that can be built, as we collectively figure out how to make AI part of our teams, part of our companies, and get it into more and more workflows across the business. I just want to ask you the jobs question right off the bat, if I could. you would think at this point there would be some data that would suggest that if AI was going to take jobs it would already be doing it and every time I speak with a company that has gone through a layoff you know the headlines are the AI has taken the jobs but they're like no no we're just restructuring our company to work in the AI era so with all these agents out in the workplace are you seeing the companies cutting down on workforce and turning that work over to agents
or do you think that restructuring answer is accurate? I think the restructuring answer is probably far more accurate. I think, look, my view is net-net. Firstly, anyone who tells you they know what's going to happen in the next five years, you probably shouldn't believe them. But my view is net-net. It'll create far more jobs than it changes. There'll certainly be quite a lot of job change. I think we focus on this binary state of lose a job, gain a job, versus the nature of a lot of jobs is changing. You know, the best parallel to this, I always think, is bank branches. It sounds like a weird proxy, but, you know, back in the 2000s, everyone said all these bank branches were going to close, online banking, mobile banking, it was all going to disappear. And actually, if you look at the last 10 years, bank branches have been growing, at least in Australia and in America. And the reason for that is it's a narrative violation. But bank branches don't do, in 2026, what they did in 2000. They do vastly different jobs for the customer. or the customer services were more important,
you changed massively what that asset is for. I think you're going to see the same thing with AI and jobs. Most people's jobs right now probably have changed a tiny bit in terms of how they get whatever their output was done. I suspect that continues to change progressively over time. So the question is, like, will we still have software developers? Yes. Will they work the same way in five years from now as they did five years ago? Absolutely not. Will we still have designers or marketers or finance professionals? The answer to all these questions is yes. Those jobs will be done very differently. Secondly, the ratio between the different job families or crafts may change, right? There's lots of views about how that's going to change. I listened to one podcast where they had a product manager basically saying he thought there'd be many more product managers and many less engineers. And the engineer said, look, I think there'll be many more engineers and many less product managers. And so the answer is, we'll see. And so I think from your answer, what you're saying is the question of will AI kill jobs is sort of less interesting.
Really, the question should be how it will change jobs. Absolutely. Yes, it's how. And I think for companies that are looking to harness AI, the question is, how am I changing the jobs that I have before some other company comes along and does what I do better because they're embracing these technologies? Right. I always love the line that I'm not afraid of losing my job to AI. I afraid of losing my job to someone who really good at using AI to do my job And I think that the way most companies should be thinking And that hard Changing the way that people think their historical priors is a very difficult challenge for most companies. It's a talent challenge more than a technology challenge. Now, again, you have access to the data about how people are actually changing the way they work. And you're selling AI products. So it's true. And you're actually at the front line of this change. So when you think about the way that people will work alongside AI agents, the way that people will change the way that their workflows exist in the AI era, what have you found?
Look, firstly, I don't like to say we sell AI products. I know that sounds weird nowadays. I think we sell technology products. I think we sell products that help companies run workflows. AI is a new ingredient in the technology mix, right? I don't say we sell mobile products. I don't say we sell internet products. AI is a very important change in the technology landscape that enables us to deliver customer outcomes, running your business processes in new and different ways. But the products themselves, I don't think we'll see the sort of, you know, the number of companies with .ai on the end of them is kind of going to disappear over time. I think the biggest change is around design and experience. Like, people are looking right now a lot about why, with all the model progress we've seen, Why are we not seeing those efficiencies flow through into company outputs, company creativity? And I would say a lot of the answer is in the design and experience. We need the talent, people to change how they think about doing their job.
And we need the product experiences and interfaces to change massively to enable that to happen as well. They've actually got to come together. Very biased, we shipped what we call our agentic multiplayer protocol, or AMP, at our conference here in Amsterdam about a month ago. And that's largely a set of experiences, technologies, and patterns on how you bring agents into your team, how they come into the interface of JIRA or Confluence or Loom or the service collection. And it's not just a set of design patterns because there's a lot of technologies underneath that. So we're rolling MCP and A to A and a whole lot of technical standards under the covers into an interface that people can use both in a familiar way, so it's very human-centric, but also it takes advantage of the capabilities of AI. And it's pretty cutting edge. I think we'll see many more interface paradigm changes in the next year ahead. Yeah, but then how does a job change? Like when you think about what the job of the future looks like, what will it look like that it looks differently than it does today?
Look, that's hard to answer generically. I think the different jobs will change in different ways. So probably the best generic answer would be people will need to think a lot more about the outcome they want to get rather than the task they're doing. So there's a little bit less of the mechanics and more about the outcome they want to get. Now, the outcome may be people talk a lot about taste and judgment. That's one part of an outcome if you're in a design or interface kind of world. If you're thinking about the outcome of a financial model, you need to think, how would I test if this model is correct? Because asking the AI to go make me a really advanced model that does X, Y, and Z, you're going to get an answer. You either trust it's correct or you need some way of verifying it, and so you're going to think far more about the outcomes of the task I'm going to do. And then I think, secondly, you're going to continually think about how do I move to a higher level of abstraction, right? If you look at a lot of the paradigms of people using agentic harnesses, It's less about ask a question, get an answer. It's more about how do I think about what types of answers I want to get
and how do I create a situation that I can run a series of loops. Those loops are going to get smarter and continue to improve and then I get towards that outcome I want to get. So sort of goal-seeking and all this type of things, I suspect, come into a lot of jobs. Now, if you need everyone to be an Andre-like engineer that thinks about goal-seeking and thinks about continual improvement and reinforcement learning, all these things, I don't think anyone's going to have that as their regular day-to-day job. Back to how do we take those ideas and technologies and put them into the interfaces of the applications that people are using every day. Yeah. You talked about how the design of the agentic flows needs to be optimized, basically, to get the maximum value out of it. Where do you see that going? look I think we need to continue to work on how agents show up in interfaces so we talk about the different modalities of human and agents working together so we think very much about chat when we talk about AI we think about
I ask a question, AI answers I ask a question, AI answers that is kind of like a one on one interaction but it's I don't think going to be the most common way of using agents If you think about using Claude to code for you, we call that delegation. So I'm telling Claude, go do this. It's going to do a bunch of stuff for me. Maybe it makes a pull request. That pull request is made by Claude or made by me or made by Claude on behalf of me. I think the last is probably the most accurate. I'm delegating work to it, but it's doing it on my behalf. When I'm interacting with an agent that's existing in the world, maybe it's on a Confluence page editing alongside me, that is really a collaborative activity. I can have two or three agents from different harnesses, different vendors. I can have two or three people. That is a true multiplayer collaboration, which is a vastly different interaction than if I talking back and forth to it And then you have more of this pairing scenario I think which is really interesting where think about it as a meeting table In the first scenario I sitting on the opposite side of the meeting table, we're discussing back and forth.
In a delegating scenario, I'm sending the agent to the meeting on my behalf. I'm giving it some instructions, I'm thinking, here's what I want you to go do in the meeting, tell me afterwards when it comes back. I think the last scenario we're starting to see is what we call pairing. So I'm sitting on the same side of the meeting table as the agent. And we're working together on the same outcome sort of together, but we're not really collaborating on it. We're almost working side by side on a given outcome. We shift a series of Loom features, for example, where that's where you see agents helping augment. If it's helping you write a document together and you're sort of back and forth co-crafting a single output, it's not quite the same as the chat GTT-like interaction. So we think a lot about the different modalities of human and agent collaboration, and then how those different modalities will show up in user interfaces because you have to understand all of this, but it has to be quite intuitive in the interface. So in Loom, that's utterly different than it is in Confluence because in Confluence, I have way more of that collaboration multiplayer kind of world where I'm editing a document together.
Whereas in Loom, I might tell the agent to go make a video on my behalf or it might send me a video that I have to watch and comment back. There's lots of different scenarios. So I think we're at the early stages of that. Maybe the summary argument is I don't believe everyone's going to sit on a command line and just interact back and forth. Yep. And how does your teamwork graph play into this process? The teamwork graph is at a different layer. So it sits underneath AMP in the Atlassian platform. The teamwork graph is an enterprise context graph we've been building for about eight years. I say that eight years ago we had maybe five people working on it. Today it's like a thousand. and the teamwork graph is trying to understand all of the objects and relationships that exist in the knowledge part of your world across lots of different contexts. So the context of things like your knowledge, so I think documents, spreadsheets, your work, projects, goals, outcomes, your communications, so email, calendars, meeting transcripts, obviously all of your source code assets, your physical assets in the building,
your data assets, so your Databricks or Tableau or something like that. And all of these come together in a single graph. We then run a lot of inference over that graph and provide the answers to that either to humans in the form of search or chat results or via a CLI or MCP to agents. So the agents that you're running, if you attach the Teamwork Graph or the Atlassian MCP server or the Teamwork Graph CLI to your codex, it will use the Teamwork Graph for far more of its queries and get far better answers. so higher quality answers which lead to lower costs, less tokens used and a faster outcome for you as well and then the teamwork graph continues to learn so every time agents query it to retrieve a fragment of the graph to do their job maybe some source code sections and maybe some documents we learn from the search results and make those search results continually smarter and we're running inference on top of those objects as well so an example of that is we can now index your agent session
so every session you're running with an agent After the session is completed, we will extract the key memories and learnings from that session, as well as all the linked documents and objects. And then those learnings will go straight back into the next session that you run via the search results. So you get a sort of compounding improvement in your agent efficacy. If your CFO is asking, it will result in less tokens used. What's very interesting about what you've just said is you actually haven't brought up the model that you're using underneath it. I have not, no. And I think your answer is very telling, and it sort of brings me back to something that I believe, which is that the models themselves are going to start to commoditize. We've seen this with things like OpenRouter, and we see it in your answer, where maybe a year ago someone would reference Claude 15 times or GPT 15 times, and you didn't. So do you believe that models are commoditizing the way that I do, and if so, what's the differentiator going to be? That's a great question. I don't think they're commoditizing maybe in that manner.
To me, commoditizing means there are so many and the net cost is basically the same that you can pick one or another. I think they will continue to have different flavors. I think they will continue to have different strengths, different types of tasks. I think they will continue to have vastly different cost profiles. I don't think they're as important as people make out. Model progress is amazing. We are seeing continued progress. Our job as a vendor is to harness that progress and deliver it in better outcomes. That doesn't mean just jumping to the latest model. That's a very expensive way to do things, and it doesn't often get better results. It's often slower. So what we call model blending is really important. So we have a lot of expertise and time spent on which model answers which question the best. And you see a lot of people like OpenRoute or others trying to do this at different scales. I think the harness is incredibly important and runs on top of the model. So you see increasing harnesses that you can sub in and out different models or harnesses that will try to make
some of that model routing happen for you for different types of tasks. And then you've got all the layers on top of that. So if you think about the context and the teamwork graph, AMP and the interfaces and the applications, I think, you know, you asked at the start, how do we see more progress in AI in business outcomes, in productivity or in the quality of products and services? I think we at that very exciting formative stage of the technology where model progress is amazing Harness progress over the last 12 months has been quite incredible We seeing a lot coming through in context progress and context efficacy and availability And then you seeing all the application stacks on top of that. So all of these moving forward at different levels, we'll see breakthroughs, I think, as the collective of the parts keep getting put together in new ways. But an interesting thing has happened recently, which is that I do think the labs have said, our models alone aren't going to be what gets us to the valuations that we're seeking. And in order to justify those valuations, we need to go upstream. So now you're seeing Claude Code, Claude Cowork, Claude Design.
When you think about the way that you work with the models, even Satya Nandela has said every time you work with a model, you're leaving breadcrumbs and they might ingest what you do and start competing with you. I would imagine that they can go into organizational software and see how companies work and maybe build a competing product. Are you worried about that? Worried is probably not the right word. I think, look, there's always this thing in technology. If you went to the 90s, it was, oh, but if Microsoft did that, they'll just crush everybody. And in the 2000s, it was if Google does that, they'll just crush everybody. And then it was Amazon. And now it's the model makers. I don't know you can run your business with the expectation that they're just going to go and do all these things because there are a lot of laws of physics that apply to every single company, right? in terms of focus, in terms of what they can really deliver. In terms of where their key expertise is, our key expertise is clearly in delivering for our customers, outcomes for their teams, their business processes, their workflows, et cetera, at massive scale with all the compliance requirements,
and back into that design and experience investment on a continual basis. That is a non-trivial thing to do. That's not an engineering technology algorithm problem. That is a more art and creativity problem. So I don't think it's likely at the large scale, but you can never say never. But at the same time, the way for any vendor to compete is just to deliver for their customers and to be very ambitious about how those things come through. So I would say the same thing in reverse. Whenever people give this argument that like, oh, isn't all software just going to be five-coded? It'll all basically be free. I'll just copy everything. because the cost of building software, the cost of writing code, I would say, more accurately, has come down, I'd say, well, maybe, except that our cost of writing code has come down as well. So it's our speed of running as a business, our speed of delivering for our customers, all of that is broadly available. So the progress that an Opus makes or an Astra makes
or Gemini, Oral, the open-weight models, Meta, Nemotron, etc., Now, that progress is available to all of the vendors pretty evenly. So it makes it in some ways quite an even playing field, which is quite fascinating. But you have to decide where to draw a line, how far to embed within an agent and how far not to. Now, a lot of software companies are going headless right now, which means basically just use the agent to access my software. I don't even need to be there. What do you think about a headless UI? Would you ever do that? You want me to say the headless is brainless again? I do think headless is brainless. I should explain this statement. If you believe the majority of your usage will be through a headless interface and therefore you don't need to continue to work on your design and experience of your software, I think that is a path to death. I think that is a very brainless move. The very idea of being headless implies you have no brain. So, you know, maybe you have a heart, maybe you have some hands, but it's not going to be very useful.
I think MCP and other interfaces, you know, we're big investors in our CLI an MCP server. We have one of the most used MCP servers on the planet at all the major labs. That is incredibly effective for how agents can interact with software, how people can write other pieces of software that use models and harnesses to access our software. I don't think that means we need to stop investing in the interface. I don't believe the majority of people will just sit at a command line or a terminal and use all of their enterprise software. It's incredibly inefficient. We built Windows and menus and pull-downs and operating system, you know, visual operating systems for a reason, we moved away from the terminal. And I think we will continue to do that. We still have a terminal. I have one on my laptop. It's very useful. It can do a lot of things, but it's not where I want to sit day to day. I don't believe users will want to sit day to day. There was a big survey that came out last week of user preferences and showed that people hated that way of working. They didn't feel it was productive. They didn't feel it was, you know,
we have a lot of visual interfaces. so I think as a if you're building software internally or as a vendor you have to invest a lot in that design experience and there are many ways that that design that experience can be adaptive with models we're seeing a lot of people playing with dynamic UI based on Jev and some of the decision models which is truly fascinating we have a whole lot of prototypes internally and things coming that doesn't mean we can move all the cheese around right like every single time you go to the UI it's totally different it's going to be very discombobulating but I do think the idea that we'll just use all software like a hidden database I think it's naive personally but we'll see that's what's exciting about the world at the moment Mike, thanks so much for coming on the show Oh man, thank you for having me Alright, thanks everybody for listening and watching and we'll see you next time on Big Technology Podcast And thanks to you, you've been a great audience Thank you very much
番組の概要欄(原文)
Mike Cannon-Brookes is the CEO and co-founder of Atlassian, the company behind Jira, Confluence, and Loom. Mike joins Big Technology to discuss how AI is changing work and the software businesses use to get it done, recorded live at Web Summit. Tune in to hear how AI agents will work alongside people, why better models haven’t always translated into productivity gains, and whether AI will eliminate jobs or change them. We also cover model commoditization, competition from OpenAI and Anthropic, and why Mike believes going “headless” is a mistake for software companies. Hit play for a grounded look at the future of work and the battle over the software behind it. Learn more about your ad choices. Visit megaphone.fm/adchoices
出典・この記事について
- 番組
- Big Technology Podcast(Alex Kantrowitz)
- 元の回
- How Humans Will Work Alongside AI Agents — With Mike Cannon-Brookes
- 配信日
- 要約作成日
- 公式サイト
- www.bigtechnology.com
この要約・解説記事は、元の音声を文字起こしした内容をもとに AI(Claude)が日本語で作成したものです。 固有名詞や数値に誤りを含む場合があるため、正確な内容は元の音声でご確認ください。制作方法について
関連エピソード

NetDocumentsのHeath Harris氏が語る「タクソノミーこそAI戦略」――最終的に勝つのはデータ
LawNext

マルチクラウドLLMルーティングと、コンテキストを理解するコーディングエージェント
Agentic Conversations (formally mlops.community)

個人創業者として戦うトークン詐欺の実態と対策
How I AI

ServiceNow投資からさらに運用価値を引き出すには――BigPanda CEO アサフ・レズニック氏が語る
The AI in Business Podcast

AIネイティブ企業が他社と違う7つのこと
The AI Daily Brief: Artificial Intelligence News and Analysis

特別レポート:ジェンスン・フアンの失墜とその後に来るもの
Big Technology Podcast