This post is based on what I tell people when they ask me about operations in AI Safety. It is a mix between “writing about my job”, writing about what operations people do, and what skills I think you should develop if you want to work in operations. I am glossing over what AI Safety entails, so if you want to find out more about the field, I have compiled some resources on my website.

TL;DR

In operations, you will do a variety of tasks to keep things running smoothly. For a thorough overview of working in ops, read this article by 80,000 hours or this one by Probably Good.
Top Skills: prioritisation, consistency, delivering good work, and never dropping a ball.
If you enjoy solving problems and helping others, ops might be great for you, and you should apply for roles!
To build the right skills, just go organise stuff. It does not have to be in AI Safety. In operations, just doing things gets you very far.
Depending on the kind of person you are, it can be worthwhile to start your own project or initiative, or volunteer in combination with building experience/gaining career capital/Earning to Give, rather than applying to operations roles.

If you want to talk more about ops, send me a message, and we can probably have a chat.

What Do You Mean By “Operations”

Operations in AI Safety means something different from in the corporate world, and your role and experience being in ops vary dramatically depending on the organisation’s culture, mission, and size. Operations is unglamorous and you will not get the applause, but you are absolutely essential for making the project run smoothly while juggling many different tasks and responsibilities (and never dropping one), and your team will love you for it.

Some common roles are:

Operations Generalist: You are expected to do lots of tasks that contribute to the org running smoothly. These roles are usually at smaller organisations and are more entry-level. They are great if you are broadly interested and do not know yet if or what you want to specialise in.
Operations Specialist: More specific roles once an org starts to grow. Usually focused on a specific aspect, like finance or recruiting.
Office Manager: For example, liaising with the landlord and contractors, sorting out repairs, fulfilling small requests, and managing snacks/food.
Executive Assistant/Personal Assistant: Making sure one person or a few specific people can fully focus on their highest-impact work.
Operations Manager: You are generally in charge of the whole or part of the org’s operations. This is usually more strategic and more management-y.

Exact tasks vary wildly, but typical tasks include: Managing money, hiring and application processes, fundraising (writing applications, compiling data), events (booking speakers, venues, logistics), legal compliance, setting up and maintaining systems and processes, office management (liaising with landlord, office supplies, bit of tech support, ordering catering), website maintenance.

That sounds like a lot, but you do not need to be proficient in all of them. The key skill is knowing when to delegate by knowing when to hire and maintain subcontractors and when to check with an expert (lawyer, accountant, etc) before deciding.

Things I Learned

You will alternate between a reactive phase, where you are just executing to meet deadlines, and a proactive phase, where you can do optimisation and reflection. Especially when you are in reactive mode, the difficult thing about ops is not any individual tasks, but reliably doing all of the tasks to a good enough standard.

You need to switch from task to task a lot, which can be overwhelming if you are unable to prioritise (either because of circumstances or because you have not developed that skill yet) or do not have a good task management system. Working as an operations generalist usually means doing a lot of shallow work, rather than deep work, and is in that way the opposite of doing research. Ops and research are very different skill sets.

How to deal with this? As an operations person, you should always make sure you are at most 80% of capacity: things will go wrong, and extra things will come up. Don’t try to push yourself to the limit. It will not work. Your brain can be buzzing with a hundred things you need to keep track of. It’s not like jobs where you have a few tasks per week. This can make it harder to switch off after work. It is easy to feel like “I just need to do this 5-minute task to unblock the team” at 5 PM, and when you check the time again, realise that it is 8 PM and you are doing ten different things. Ops burnout is not like researcher burnout, where the one looming deadline and lack of feedback pushes you over the edge. It’s burnout by a thousand cuts.

People do not comment on things that run smoothly, except if you solve a problem that people have already noticed. No one will profusely thank you for managing food orders without issues, but you will definitely hear it if you forgot about one dietary requirement and someone now has to subsist on breadsticks. Good ops looks like “business as usual”. Combined with the fact that in small organisations, you mostly work alone, you need to be able to internally recognise your value more than in other jobs.

What Skills Do You Need

Essential

Proactive: Do not just wait for someone to ask you to do something. If you see something that you can improve, just do it.
Execute reliably: In ops, you are executing a lot. If you are asked to do a task, will you be reliable and figure out a way to do it, including unfamiliar ones? You will do many things you have never done before, and that is part of the fun, but it also means you need to learn quickly.
Communicating well: You need to communicate with lots of different stakeholders (researchers, contractors, landlords, applicants, funders, your team) about the value and feasibility of certain plans so that they can make better decisions.
Attention to detail: Are things you present clearly formatted, have you included all the right information, have you completed all the steps you need for a certain task? Checklists are your friend!
1. Related skill: Anticipating problems before they occur.
Doing good work, but not being so perfectionist that you don’t get everything done that needs doing: 80% of the result in 20% of the effort is usually good enough, but not always, and being able to know when you need to put in 100% of the effort for the 100% of the result is important.
Prioritisation: People will throw many things at you, and they will all sound important. You need to be able to determine what is actually important, and do those first.
Connect ops to strategy: Great ops takes into account what the organisation’s goals are, and how you can create systems to achieve those. Consider how you set up yourself and the organisation to execute your strategy well.

Important

Using systems: It is not important to be naturally organised to be good at ops. It is much more useful to be good at working with a system like task management software.
Don’t mind doing repetitive work sometimes: Some things cannot be automated (yet), so you will need to do them manually. Whether it is looking over transactions, sending slightly personalised emails, etc. Invest in tools to make it easier, but accept that this is part of the job, or even any job.
Optimisation mindset: Can you find a way to save your future self time or other people in your org time? Make sure to deeply understand the problem here, rather than just applying whatever solution you already have in your head.
Be comfortable enough with spreadsheets that they do not make you or other people cry: There have been too many times when some nice conditional formatting and an XLOOKUP has made people’s day. LLMs are very good at helping with this.
1. A related skill here is understanding databases and how to structure data. You do not need a computer science degree, but understanding what kind of relationships (one-to-one, one-to-many, many-to-many) exists and how to best split up your data over different columns comes in handy.
Delegation: What can you hand off to other people, contractors, LLMs, etc?

Sometimes you also need subject-specific knowledge (finance, people operations, etc), depending on the role, and leadership skills if you are in a management position.

You Do Not Need

Academic credentials: You just need to be able to do it. Do not get an MBA. The main value of an MBA is networking, but this is much easier and cheaper to do in AI Safety in other ways, such as going to an EAG(x) conference.
Domain expertise, such as technical knowledge of AI: It is helpful to vaguely understand the work the organisation does and why it is important, but you do not need to be able to understand a paper.
Mass communication skills, such as public speaking, writing for external audiences: You will need to coordinate and communicate clearly within your team, and in a small organisation, you might need to write some public copy for a website, event, etc, but you do not need to be an expert.
Strong charisma: It helps if your team likes you, but you are mostly coordinating behind the scenes.
Being naturally organised: It is much more important to be reliable and good at using systems.

Building Those Skills

You just need to do things. After you have done it, reflect on what went well and what did not go so well, and how to improve next time. The learning is in the reflection, not just the doing.

What can you do? Well:

Organise events: Whether it is a party, meet-up, speaker event, conference, performance, reading group. It will teach you a lot about planning and logistics
Make yourself a budget or another kind of spreadsheet to track data or make decisions with.
Use task management software to organise your personal life. I recommend looking at the basic principles of Getting Things Done.
Join or start a student society, like an AI Safety group, and run their operations
Intern or volunteer for an existing organisation
Start your own thing. This requires higher agency, but can be faster than finding a role in an existing organisation. It can also be much more impactful, though not by default, if your initiative scales and fills a gap in the ecosystem.
1. For some tips on deciding what to do, check out “Focus on the places where you feel shocked everyone’s dropping the ball” and “Deciding What Project/Org to Start: A Guide to Prioritization Research”

You can find more on this topic by reading So you want to do operations [Part two] - how to acquire and test for relevant skills

How To Get A Role

There are many reasons you can get rejected (bad application, bad test task, bad interview, not a good fit with the specific role or organisation, someone else was simply more qualified, etc, where bad can be because of various factors that you can improve), but these are some ways you can set yourself up for success:

Have some AI Safety context. I have compiled some introductory resources on mickzijdel.com/ai
Read about what ops roles entail (like you are doing right now), so you know what skills hiring managers want you to prove you have in your application, and so you can anticipate what you might be asked in the interview
Have some demonstrable experience. In my experience, hiring managers are primarily looking for an ability to get stuff done effectively and an ability to learn new skills, no matter the context, rather than previous experience in operations. This is more true for entry-level roles than management roles. Because of the generalist nature of these roles, it can be a boon to have had many different kinds of previous experience.
Do a good test task. Being proactive, knowing which parts to prioritise, and communicating well are all part of this.
Volunteer or get involved with AI Safety in other ways, so people get to know you and your work

How fast you can transition into AI Safety operations depends on your previous experience. If you are looking for an entry-level role and have previous experience in organising, you just need to build some context. If you lack any experience, it might take a couple of months to a year to build that. If you are aiming for more senior roles, the requirements become more org and role-specific, but you will need to show past outputs, likely need more AI Safety context, and most likely need management experience.

For further reading, Laura from 80,000 Hours’ post on finding an impactful career when the job market is not cooperating, Gergő Gáspár’s post outlining why he thinks experienced professionals struggle finding roles in AI Safety, and Eli Nathan’s post on why you might be getting rejected from (junior) operations roles are all great resources.

How I Got Into AI Safety Operations

In high school, I organised concerts and got into productivity tools, and while at university, I took part in theatre productions (budgets, logistics, team management) and ran our theatre building. It was chaotic and there were many stakeholders (other committee members, company members, the university, the Students’ Association, the public), but we did a show every week so things just had to be done.

Every summer, I tried a different job (in order: software engineer, acoustic consultant, physics teacher), but none of them had the “run around and solve problems” part that I liked about theatre. I recommend trying out many different jobs/internships while at university to find this out for yourself. You can just email people to ask. In the summer after I graduated high school, I emailed a few software development companies in my town and one of them offered me a job for the summer because I had some programming experience, and they asked me back the following summer. The summer after that, I emailed a building consultancy firm in the Netherlands explaining that I was studying acoustics, and asked if I could work on one of their acoustics teams for a few months. Both of these were big firms, so they had the capacity to take on an extra person and give them mentoring and a small salary, but they were not so large that you get lost in the bureaucracy. The Physics Teaching internship was through the Ogden Trust’s Teach Physics scheme.

After graduating, I started working for my university’s Students’ Association to support the 370 student groups at Edinburgh. Again, I found myself doing ops work like managing training programs, writing guides, answering lots of emails, helping with small requests, and improving systems and processes, which I really enjoyed.

As I got more concerned about risks of advanced AI, I decided to quit and look for an AI Safety role. I was rejected many times because the roles I applied to were not a good fit for my background, but I ultimately found my current role after I went to my first EAGx conference and heard about the role there. I think the largest factors in me getting the role were the diversity of past experience, being able to do “good operations work” in the test task, and actively going to a conference and meeting people to ask about their jobs.

At the time, I was also considering starting my own project, but I preferred the guaranteed chance of skill-building and making connections at an existing organisation. If I had kept being unsuccessful at applying, this is what I would have done instead, and it is something I am still keeping as an option for the future.

What Do I Do In My Job?

I work as Program and Operations Manager for Catalyze Impact, an AI Safety incubator. Here is a non-exhaustive list of things I have worked on in the past year:

Managing the application process for our incubation program:
1. Setting up a system to track applicants, reviews of their submissions and their progress through the process
2. Setting up a system to score applicants on their applications, test tasks, and interviews
3. Interviewing candidates
4. Sending outcomes and feedback
Logistics for the program
1. Communicating with participants
2. Writing a program handbook
3. Liaising with the venue
4. Inviting and liaising with speakers
5. Organising a mixer
6. Processing payments and reimbursements
7. Setting up a Slack workspace
8. Devising a process for cofounder matching
Marketing
1. Promoting the program on social media and Slack groups
2. Individually reaching out to referrals and candidates in our network
3. Managing the website
4. Representing Catalyze Impact at conferences
5. Designing some graphics
Impact evaluations
1. Figuring out how to track our core metrics
2. Designing feedback surveys
3. Following up on people to fill out the feedback surveys
Systems Improvements
1. Automating repetitive tasks in our Airtable
2. Setting up new Airtable tables and bases to track data
3. Making sure we use standards for our Airtable
4. Keeping our Google Drive organized
Compliance
1. Data protection
2. Liaising with our fiscal sponsor and accountant about our financial records

A lot of different things, and a lot of them I had not done before. For example, this was my first time using Airtable, but I figured it out quickly because of my previous experience with spreadsheets and databases, and now I feel extremely comfortable using it.

Appendix: Other resources

The comments on these posts are also valuable because they provide extra context and other perspectives, so do not skip them.

Top Recommendations

Experiences

Some of these can give an outdated take on the jobs landscape. The posts I link below are more up-to-date.

Finding An Impactful Role

From Comfort Zone to Frontiers of Impact: Pursuing A Late-Career Shift to Existential Risk Reduction (not about ops directly, but I appreciate Jim honestly sharing his experiences finding a role as a late-career professional transitioning to operations roles in AI Safety)
1. Appendix A is a comprehensive general feedback email on a Operations Manager position, with very useful tips
How to have an impact when the job market is not cooperating [EAG Bay Area + EAG London workshop]
Why experienced professionals fail to land high-impact roles (FBB #5)

How To Be Better At Ops

Read Getting Things Done and set up a system
Resource: readings to learn more about HR/people operations work
Resource Portal | Anti Entropy
Resources | Impact Ops

Operations in AI Safety: A One-Year Perspective and Advice