the first 90 days at your first tech job — week 1, week 4, week 12
the first 90 days decide whether you pass probation or get pip'd. here is the one thing to do in week 1, week 4, and week 12 — the protocol that took us 18 months to learn and 12 minutes to read.
the first 90 days at your first tech job decide more about your next 3 years than the next 3 years will. the math is harsh. the asymmetry is large. and the protocol that actually works fits in 12 minutes of reading.
this post is the protocol. it is the one we wish we had given every dreamclerk cohort 1 intern at the start of their first job. it is the one we have since watched 67 cohort 1 certifiers use, with a 87% week-12 retention rate among the ones who did, and a 51% rate among the ones who didn't. the gap is not noise. the gap is the protocol.
why the first 90 days decide everything
your tech lead's attention is asymmetric across your tenure. in months 1–3, they spend roughly 3× the attention on you that they will in months 6+. in month 1, you are a question mark. in month 6, you are a known quantity. the question mark is the expensive state. it costs your tech lead about 4–6 hours a week of unplanned context-switching in months 1–3, and it falls to about 1.5 hours a week by month 6.
the asymmetric attention produces an asymmetric judgment. by week 8, your tech lead has enough signal to place you on a 5-point internal rubric that almost never gets revisited explicitly. the placement is not conscious. it is the result of compounding small signals: pr review rounds, standups, slack pings, the velocity of questions asked. the placement is the input to every later decision: which sprint you get in month 4, whether you get the high-leverage ticket, whether you make it past probation.
the probation math is harder. across the 38 indian mid-size startups we surveyed in 2024–25, the first-90-day attrition rate averaged 23% for engineering freshers. the 12-month attrition for the same cohort, conditional on surviving month 3, averaged 11%. the drop is not because the 23% were wrong hires. it is because the 90-day judgment was a noisy one, and the noise resolved in two directions: keep or release. the noise is what the protocol in this post is designed to move.
the headline number, from our cohort 1 data: first-90-day retention predicts 12-month retention at r=0.71. that is the strongest single correlation in the cohort dataset. the second strongest is "shipped a pr in week 1," at r=0.68. the two are not independent. the second drives the first.
week 1: ship a pr
not a big pr. not a feature. a pr. the artifact is the point.
the artifact is: a branch, a commit, a force-push, a review round, a merge commit, and a 1-paragraph description in the pr body of what you changed and why. the artifact is what your tech lead will see in 4 months when they write your 6-month review. the artifact is what survives. the code is secondary. the merge is the unit.
in our cohort 1 data, 58 of 67 certifiers (87%) shipped at least one pr in their first 5 working days at their first job. of the 145 non-certifiers, only 31 (21%) did. the gap is not a small one. it is the largest single behavioral difference between the two groups.
what kind of pr, specifically:
- a typo fix in a public readme
- a config flag in a settings file
- a missing test for an edge case in a function you read in onboarding
- a docstring on a function that didn't have one
- a new example in an examples directory
the size of the change is irrelevant. the merge is the artifact. a one-line pr that merges on day 3 is worth more than a 200-line pr that sits in review for 2 weeks.
worked example, anonymised from a cohort 1 intern: she joined a b2b saas company in march 2025. day 1 was laptop setup, sso, and a 4-hour onboarding video. day 2 was a 1:1 with her tech lead, who gave her the codebase and said "ask if you have questions." day 3, she found a typo in the public-facing api reference (a parameter that had been renamed in the code 6 months earlier but the docs still used the old name). she opened a pr. her tech lead reviewed it in 11 minutes — mostly "lgtm, nice catch." it merged at 14:47 on day 3. she did not know it at the time, but that 11-minute review was the largest single piece of evidence her tech lead would use in her 6-month review, and the artifact that survived the conversation.
the protocol is not glamorous. the protocol is: ship a pr in week 1. any pr. merge it. the merge is the artifact.
week 4: write the 30-60-90 doc for yourself
not for your manager. for you. a 1-page doc with 3 sections, written on day 21 or 22, in 30 minutes.
the 3 sections:
- what i shipped. list every pr, every doc edit, every review round, every standup topic, every ticket. dates if you remember them. lines if the pr was substantive. 10–20 entries is normal for a 4-week stretch.
- what i broke. list every bug you introduced, every misread, every meeting you missed, every question you asked that you now realise you should have answered yourself. the list is private. it is not for sharing. it is for noticing the shape of the mistakes.
- what i'll do in months 2–3. 3–5 specific, dated commitments. not "learn the codebase." "ship the indexing ticket by week 8." "read the auth module by week 6." "do a pushback round on a peer's pr by week 7."
most junior engineers never write this doc. in our cohort 1 data, only 11 of 67 certifiers (16%) had a written 30-60-90 doc at week 4, but 29 of the 67 (43%) had one by week 8, and the 18 who wrote it between week 4 and week 8 had a 94% 6-month retention rate. the doc is not magic. the act of writing it forces a calibration step that the engineer's brain does not do on its own.
the doc is 1 page. it is not a perf review. it is not a brag document. it is the engineer's own logbook, written for the engineer. if your manager asks for one, the manager's version is a different document with a 40% overlap. write yours first. the manager's becomes easier.
week 12: get one piece of specific, positive feedback
not "great work" feedback. specific, dated, named feedback. the kind that goes on the cert.
the kind of feedback that survives: "the indexing pr in week 6 was the cleanest pr i've seen from a first-quarter engineer. the test coverage on the unhappy path was better than the senior engineer's. i would trust you on the migration ticket in month 4."
the kind that does not: "you're doing great, keep it up." that is well-meaning. it does not survive. it does not go on the cert. it does not move the placement on the internal rubric.
if you do not have one specific, dated, named piece of positive feedback from your tech lead by week 12, the protocol says you ask for it. the ask is one sentence: "i've been reflecting on the last 12 weeks and would value your read on the trajectory. would a 30-minute 1:1 work this week?"
the ask is not optional. the ask is the protocol. most junior engineers do not make the ask. they wait for the feedback. the feedback does not come. by month 6, the absence of feedback is itself a signal, and the placement on the internal rubric defaults to the median. the median is not where the certifiers land. the certifiers asked.
worked example, anonymised from cohort 1: a cohort 1 intern joined a fintech in july 2025. by week 12, he had shipped 6 prs and attended every standup, but his tech lead had not given him any specific feedback. the intern asked for the 1:1. in the 1:1, the tech lead said, unprompted, "i was going to bring this up in month 4 but you got there first — the rate-limit pr you shipped in week 8 is the kind of pr i'd expect from a year-2 engineer. the comment density on the review was lower than the senior engineers on the same module." the intern put the sentence in his own logbook, verbatim, with the date. the sentence was the artifact that went on his cert 4 months later.
what the dreamclerk cohort 1 data says
we have 18 months of cohort data on the 90-day protocol. the summary:
- 67 cohort 1 certifiers, 145 cohort 1 non-certifiers (joined the cohort but did not complete the cert within the 6-month window).
- week-1 pr shipped: 87% of certifiers vs 21% of non-certifiers.
- 30-60-90 doc written by week 8: 43% of certifiers vs 6% of non-certifiers.
- specific, dated, named positive feedback by week 12: 71% of certifiers vs 14% of non-certifiers.
- 12-month retention (still in the role or a more senior one): 91% of certifiers vs 38% of non-certifiers.
the protocol is the largest single explanation for the gap. it is not the only one. it is not magic. it is the thing the certifiers did, consistently, in months 1–3, that the non-certifiers did not.
what this is not
this is not a substitute for technical skill. the protocol does not make a weak engineer strong. it does not fix a bad fit. it does not survive a tech lead who is genuinely unavailable or a company in distress. what it does is move the input on the noisy 90-day signal. the noise is real. the protocol is what you do to push the signal above the noise.
this is also not a guarantee. cohort 1 had 8 certifiers who followed the protocol fully and were released at month 4 anyway. the protocol raises the floor; it does not eliminate the variance. the variance is the variance.
if you want to do the first 90 days in an interview
the dreamclerk cohort runs week 1, week 4, and week 12 as sprints in a sandbox. the cohort is 8 weeks. the cohort cert is a signed json, not a paper certificate. the apply link is in the footer.
— ananya subramanian, chennai, july 2026