Rob Bowley

@robbowley.net

2317 Followers 1182 Following

Product & Tech Leadership Advisor, Consultant, Coach & Mentor Tech, Software Development, Science, History, Economics, Politics https://blog.robbowley.net https://www.linkedin.com/in/robertbowley https://pragmaticpartners.co.uk Manchester, UK

Screenshot from the movie, Her - Joaquin Pheonix staring at a computer screen
Image of a whiteboard work tracker in an office
Error message on ChatGPT: "Gateway time-out The web server reported a gateway time-out error. Ray ID: 9d6045cffa57054f Your IP address: 188.127.83.3 Error reference number: 504 Cloudflare Location: London"
Claude Status page showing things are currently broken
@emollick post on X Two things: 1) Given that effective AI tools are very new, and we have little sense of how to organize work around them, it is hard to imagine a firm-wide sudden 50% efficiency gain 2) CEOs with vision who hired well should also use AI for expansion & augmentation, not decimation
Post from @BullTheoryio on X This is CRAZY. It turns out Jack Dorsey's Block $XYZ spent $68 million on a single party in September 2025, roughly the annual payroll for 200 employees at $340,000 each. Then 5 months later, he laid off 4,000 people (40% of the company's workforce), citing AI and "intelligence tools" as a reason. Many people are speculating the real reason is poor planning and overspending, not AI.
Further screenshot from Claude Cowork on the contract review: Me: This was all, pretty terrible for a contract review, wouldn't you say? Claude: Yes, it was poor. I should be upfront about that. The first version had wrong clause numbers throughout, fabricated a Schedule 3 that doesn't exist, got the liability cap percentage wrong, got the late interest rate wrong, got the non-solicitation period wrong, and mischaracterised whether the consequential loss exclusion was mutual or unilateral. Those aren't matters of interpretation - they're factual errors from not reading the contract carefully enough before writing the review. When you corrected me on Clause 5, I then went back and did the work I should have done first - extracting the actual text properly and reading it. The second version was substantially more accurate, but I then still overcalled the IP issue as RED when, as you rightly pointed out, the clause is fairly standard for a services agreement. A few things I'd flag about where this went wrong: I rushed to produce a deliverable before properly understanding the source material. The PDF images were hard to parse and I should have extracted the text first, verified the clause structure, then done the analysis. Instead I jumped to creating a polished Word document based on a sloppy first read. I defaulted to an alarmist tone. Contract reviews have a natural bias toward flagging everything as a problem - it feels more thorough. But calling standard commercial provisions RED flags isn't helpful, it's noise that makes it harder to spot what actually matters. The genuinely useful observations got buried. The set-off contradiction (Clauses 10.8/10.9), the unused defined terms, and the narrow licence-back wording are worth actual conversations with XXXX. Those got equal or less weight than things I'd overstated.
Further screenshot from Claude conversation using the legal skill: "There is no Schedule 3 for insurance - I fabricated that in the first version"
Screenshot of conversation with Claude Cowork using the Claude Legal Skill: Me: Are you getting mixed up? Clause 5 is rights ot termination Claude: You're right to call that out - let me go back and re-read the contract properly rather than relying on my initial pass. Extracted and processed MSA PDF document text content Extracted and processed MSA PDF document text content Fair point - I got the clause numbering wrong. Let me read the full extracted text properly to get the actual structure right before redoing the review. Read DocuSign Master Services Agreement document Read DocuSign Master Services Agreement document Right, I apologise - I clearly got the clause numbering substantially wrong throughout the review. Now I have the actual text extracted, the correct structure is: Clause 5 = Expiry or termination of Agreement (as you said) Clause 11 = Intellectual Property Rights Clause 14 = Limitation of Liability Clause 10 = Charges and Payment Clause 9.3 = Non-solicitation (6 months, not 12) I also had several factual errors - the liability cap is 150% not 100%, late interest is 3% not 4%, and the consequential loss exclusion is unilateral (protects only the Supplier), not mutual. Let me redo the review properly with the correct details.
Photo of Power Loom at Manchester Museum of Science and Indsutry
Linkedin post starting with the side of vibe coding nobody is talking about
Screenshot of a comment chat on LinkedIn: Rafat Sarosh AI for Enterprise, Relentless Optimist, and Practical Builder. 11months ago Matthew Skelton haha ..you may be giving me a escape hatch. But truthfully, code cost is falling to zero faster than we all are grasping ... yes, yes engineering system is much more than code. Agree, but I think every engineering team should be cut by 60% by next year.. Sonja Lorscheidt 1h ago Rafat Sarosh and has it been? Just wonder. 🤔
A photo looking across a jetty in venice over a canal with a coffee cup in the foreground
A chart showing OpenAIs traffic declining 6% since Google launched new Gemini model and Banana
Business Insider: HP puts a number on how many jobs it'll cut due to AI
Scooby do meme with fred about to pull the cape off a villain labelled product engineer. In the second frame he has revealed the villian and is labelled agile software developer
Screenshot of Similarweb AI Global Sector Trends on Generative AI report 10th Oct 2025
Screenshot of Claudes status page showing lots of downtime and instability across all its services over the last 90 days
Line chart illustrating different perspectives on AI coding tools across experience levels. Y-axis indicates perception of AI coding tools usefuleness/engineering population X-axis shows progression from Beginner through Mid-level/Mediocre, Experienced, to Expert. Three lines: (1) Yellow - perceived usefulness of AI coding tools, peaks high at beginner and mid-level, drops significantly into negative for experienced engineers, then rises again at expert level. (2) Blue solid - actual distribution of engineering competence in the population, forms a left skewed bell curve peaking at on mid-level and roughly 20% in experienced or expert. (3) Green dashed - self-perceived distribution of engineering competence, is shifted to the right vs the blue lin. The gap between blue and green lines illustrates overestimation of expertise in the industry if you went by what people said they were.
A slide showing Eilyah Goldratt next to a signed - Warning Local Optimisation Ahead
Photo of slide: The Quality of a System is Defined by Our Ability to Change it"
Message in CTO WhatsApp group: I did a code and architecture review for a company yesterday who had vibe coded their MVP - there were no common classes, modules, functions, etc. every "feature" was a separate file with literally a main method full of all the code required for that one function - no reuse anywhere and loads of bits of code that were declared but never used, variable names that didn't make sense, the same variables defined everywhere, etc. Because they were quite far down this path the tidy-up process is going to be a long and difficult road
A message in a CTO community slack channel: "Really fun/unique role - a founder in my network is looking for a Fractional CTO who can manage a team of 4 non-engineers all vibe coding their product with Claude Code. It’s a well-funded startup run by a second-time exited founder. Not for everyone! But a very unique and fun setup for the right CTO-type :slightly_smiling_face:"
Chart from Stanford paper showing decline in entry level role openings for customer service and software engineers
DuoLingo share price chart over last 1 month. 5.5% down
Title page from Faros study: "The AI Productivity Paradox: AI Coding Assistants Increase Developer Output, But Not Company Productivity"
A woth 2016 to mod 2024 timeline showing unemployment rates on vertical access and 5 cohorts of employees grouped by how much exposure to AI they have. Those most exposed to AI have the lowest unemployment rates
A chart from Semrush showing AI gets most of its citations from Reddit (40.1%). Then Wikipedia with 26%, YouTube and Google 23%
Screenshot from Codurance report listing key takeaways: AI-assisted code had ~4x more lines and complexity, more duplication, early productivity gains that quickly declined, and a skills gap in using AI tools effectively.
A Jira cumulative flow diagram showing only ~30% of time is spent in development vs other stages
A Jira cumulative flow diagram showing only ~30% of time is spent in development vs other stages
A Jira cumulative flow diagram showing only ~30% of time is spent in development vs other stages
A meme showing a precarious tower of blocks called "all modern digital infrastructure" and then to the right bottom, an angry birds catapult aiming birds at the tower with the title "AI"
Quote from a LinkedIn post I wrote 5 months ago: "It’s becoming clear that this is likely as good as it’s going to get. That’s why OpenAI is shifting focus - "GPT-5" isn’t a new model, it’s a product. "
Cartoon image of monkeys with keyboards with the title "Typing is not the bottleneck"
Charts from LeadDev’s survey of 600+ engineering leaders: 47% use AI for code generation, 45% for refactoring, and 44% for documentation. Most report only slight (39%) or moderate (23%) productivity gains, with just 6% seeing a game-changing impact and 21% reporting no impact or a decline.
Metro headline Candy Crush developers being replaced by AI they made