Streamlining Technology News for Schibsted's Engineers
Otherwise known as "The Schibsted Tech News Weekly"
Blog posts, CVE announcements, library change logs, Twitter feuds. How does anyone keep up without endlessly doom scrolling? How do you help engineers in a large company stay up to date? Or a better question: how do you help keep engineers in a large company stay up to date with the relevant news stories?1
There’s so many different topics to stay on top of2, especially in the world of technology. Engineers will typically follow news that involves their particular niche. React developers will follow React releases, iOS developers will follow iOS news. However, it’s possible to be an expert in one topic while also having a broad knowledge in other topics. The concept of a T-shaped individual is pretty handy for visualising this: the base of the “T” represents their expertise, and the roof of the “T” represents their knowledge across a wider range of areas. To be a future-thinking tech organisation, we’d like to follow this pattern: developers who are extremely skilled at one particular area, but with capabilities and insights into other areas3. A very easy example of where this is useful is frontend development: you can be deeply invested into one framework, but having knowledge of other frameworks may lead you to better solutions.
How do we spread knowledge, and strengthen both the expert skills and the generalist skills? We’ve got a history in Schibsted of doing this, with The Schibsted Daily:
Every day several thousand employees receive The Schibsted Daily, with short and relevant news from the media business, targeted and curated for employees in Schibsted. The purpose is to share knowledge and at the same time contribute to building an innovative and forward-leaning company culture.
We’ve applied that model now to tech news, where once a week we will post in our tech channel 4 interesting news stories from the last week. There’s a summary for each article, along with some context on why it’s relevant to us.
I initially thought of the concept when reading about the upcoming iOS changes in the EU due to the DMA, which actually will affect all our user-facing developers (frontend and mobile). There was a burning question in my head, “how do we make sure everyone is aware of this and how will it impact us?”. Simply sharing the link will just add to the noise and news fatigue, and anyone not into Apple-related news would probably skip it. The announcement and developer page together were about 15,000 words, with plenty more in the comments on HackerNews and other tech sites. By digesting the important parts, I could write a summary that was relevant and provided context.
Since I was already doing it for that article, I thought why not include some of the other articles I had read that week. My chosen platform for sharing this message was Slack, so I used a screen’s height at the default resolution of Slack on desktop to guide the writing process: a summary of 4 links each composed of title, summary, and the context would typically take up one screen’s height. It’s not a perfect measurement, but I wanted to avoid the readers from having to scroll too much on desktop. On mobile, it’s possible to react or reply to a long message by holding down on the message. On desktop, you have to scroll to the top of the message to reply in the thread if there’s not already a reply thread. Additionally, we didn’t want to make this too long: the tech news weekly should provide signal and not noise. The longer a message, the more likely it is to be considered noise.
The first Tech News Weekly went out pretty well - with a bunch of positive feedback. So we figured we’d run it again, once a week.
In media, we don’t do anything without data to tell us how useful or valuable it was. So how do you track analytics on a Slack message? A low effort solution is to get emoji reactions. So every title of each summarised article begins with an emoji, and readers are asked to react with that emoji for whichever articles they liked. We call the combination of news content topics “the news mix”, and these reactions help fine-tune the mix of what news is covered.
It wouldn’t be a data-driven ambition without a Google Sheets backing it up, so we made a Google Sheet that tracked the date, the link, the topic, the title, the summary, the context, and the feedback. From that we calculate the length, to figure out when something is too long or short. A short Apps Script function then pulls all of those together for a specific day, and puts them into a Slack-friendly Markdown comment. There’s also a function that will fetch stats and publish to the Slack channel, though for now I manually post the message — it’s a lot easier to catch any errors on your final read-through before posting.
The default news mix is based on the following criteria:
Updates on tech we use for infrastructure, libraries, and programming languages.
Industry news that will affect the organisation.
New approaches and techniques for infrastructure and programming.
Advice on how to use tech in different ways (e.g for performance or efficiency).
Avoid AI. There are already so many discussion forums for it.
This will evolve over time, but it’s important to have a variety: we don’t expect every article to be interesting to everyone, but if we manage to cover something everyone is interested in each week, we’re doing a good job.
Articles are discovered mostly through link-aggregators, such as HackerNews or Lobste.rs. It’s important to remember that the goal is provide relevant content to the readers, so we don’t just pick the top 5 from the last week, and instead filter through to see the ones we think our organisation should care about. Often the comments on these sites help provide context on why it is interesting, particularly when it is not clear the impact that a version release will have. Twitter used to be a good place to find content, it’s not any more. Much of that news has moved to LinkedIn.
To be aware of what’s relevant, we have to keep track of what our teams are using, and what they’re interested in. There’s not much point sharing a JQuery minor release if all your teams have moved on to more modern frameworks. We do this mainly through having a couple of forums where the tech leads or architects can share and discuss what their teams are up to. Some teams will discuss on their open Slack channels what they’re doing, or even give open presentations. All these sources provide valuable feedback on the directions teams have diverged from, are currently taking, or planning to take in the near future. To successfully influence a system, you must first understand the needs, the history, and the actors.
The next steps will probably be to find the right level of detail. Are 3 articles covered in more detail more valuable than 4 articles in less detail? Should each newsletter have a theme? What should the tone of voice be? We’ll also consider if we should reach the audience in alternative ways, such as an email instead of a Slack message.
We’ll lead into discovering the technology trends of the company through a tech radar. There’s considerable overlap between the Tech Weekly and creating a tech radar: they both require good insight into the context of the company.
It’s been a successful endeavour so far, and we’ll continue iterating as long as it continues to provide value! I’ve thought about potentially creating a more public version, top 4 news stories of the week, briefly summarised and sharing why we as a media company would care about them. But I’m not certain if that’s particularly compelling for anyone outside of Schibsted. I’d highly recommend repeating this in your company - it usually takes less than an hour each week, and can provide a good return on that time investment. If nothing else, it inspires conversation.
Setting up your own internal newsletter
These steps roughly outline what we did to set up the Tech Weekly4.
Figure out who you want to target and why. Why does this newsletter need to exist? To whom does it provide value? Who is your target audience?
Identify how you will reach the audience. Slack, email, other? How will you track reactions or other analytics?
Describe your news mix. What ratio of topics should you have?
Identify sources and places to pick up articles from. Is there enough content to do the summary at the frequency you intend? Every time you find a new source, add it to the list of places to check.
Write a writing style guide. Is your tone of voice friendly? Do you use emojis and jokes? Is it purely factual? How many articles per newsletter will you cover? How long should each summary be?
Document your routines: on Thursday do X, on Friday do Y. Define your success measures based on the analytics you are able to collect from your platform.
Set up a spreadsheet with all your articles and stats. Set up scripts to automatically5 give you more stats on your content.
Launch, evaluate, reassess. Scale down if it’s not working.
If you’re curious about how we’ve set up our Google Sheets and Apps Script, feel free to reach out to me at LinkedIn.
Should we use AI for summarisation?
The summarisation could be done with AI, but currently it’s done by hand. Creating a summary requires the writer to be aware of the topic, understand and filter the article contents and comments, and pull in any historical context. This is then combined with the context of the organisation to ensure that the output is concise and useful. With the right prompt, this is definitely doable by AI today. There are a few reasons why I haven’t jumped to AI:
I generally read a lot of tech news stories anyway, so it doesn’t take much effort to summarise or find stories.
Being familiar with the news stories myself helps in my day-to-day discussions and work.
The human brain is amazing at linking contexts and seeing a picture beyond current AI capabilities.
If I were the reader, I’d rather read a human’s perspective (and opinion) on a topic rather than AI’s.
Just to illustrate the point, here’s what ChatGPT gave summarising what aspects of the DMA changes to iOS would affect a news company with websites and apps in the Nordics in 2 paragraphs:
Whereas my take looked something like this:
The ChatGPT one is full of business big words which sound fancy but mean little. My hand-written one stated the impact and the scope. Short and to the point.
With some prompt engineering, you could definitely get ChatGPT closer to the hand-written version. It’s definitely an option in the future, but for the reasons listed above, we’ll stick with the hand-written version for now.
Schibsted Media is a news company in the Nordics with some of Sweden and Norway’s most popular news destination. We’re undergoing a technical transformation, driven by a reorganisation. As part of the reorg, the CTO Office came into existence to help the organisation change and grow. My role is a tech enabler in that office, working with the engineers to facilitate and enable them. Our technical organisation has around 250 engineers or people in engineering adjacent roles, spread over 3 countries.
News fatigue typically applies to news that has a strong emotion reaction (e.g lots of heavy headlines), but it can also apply to being overwhelmed by the quantity of news in general. News avoidance can happen as a result, where readers will no longer seek out or read stories. Our hypothesis is that readers will end up ignoring links if a summary is not provided, such as those in busy channels where many links are shared each day. Our remedy is to provide those summaries whenever sharing a link.
The “T-shaped individuals” concept is usually used to describe people with cross-functional experience, but it can also be used to describe any kind of specialist with a wider experience.
I try to document as much of what I do as possible - either through a personal “win log”, or through writing playbooks. A playbook does not need to be complicated, but can simply state your goals and ambitions, followed by how something will be carried out.
If you have to do something manually on a regular basis, look into automation to cut out all the boilerplate so all you have to focus on is the core concept.