/Tech1d ago

AI Helps Start Research Projects But Struggles To Finish Them

22148153531.7K
John B. Holbein
Chenhao Tan#604
Original postChenhao Tan#604
John B. Holbein@JohnHolbein1

I was thinking this morning about how AI is pretty good at starting research projects, but not very good (at least for now) at finishing them.

As a result, it’s become much easier for scholars to accumulate an unmanageable backlog of half-baked, AI-assisted projects.

Finishing research has always been the hard part. But I’m not sure our PhD training and professional development institutions have fully adjusted to a world where the gap between starting and finishing has grown exponentially.

7:19 AM · Jun 9, 2026 · 31.7K Views
Sentiment

Users counter the claim that AI struggles to finish research projects by reporting success with tools like Claude as coaches that help complete papers and manage idea routines.

Pos
100.0%
Neg
0.0%
2 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS564LIKES11
Ryan Briggs@ryancbriggs

@JohnHolbein1 "I was thinking this morning about how AI is pretty good at starting research projects, but not very good (at least for now) at finishing them." ← PhD-level intelligence

1dViews 564Likes 11
RETWEETS1
Adam David Long@adamdavidlong

@JohnHolbein1 @mattBernius Maybe I’m just repeating your point but — for myself I’m noticing AI makes it super easy to start project B once project A hits the “hard part” ie actually finishing it.

1dViews 11Likes 1
REPLIES1
Adam David Long@adamdavidlong

@JohnHolbein1 @mattBernius Im also using Claude as a chief of staff / coach and have had some success with a routine where it writes down idea for new project but then says “we can do that once we finish this current project”

1dViews 8
joseph francis@joefrancis505

@JohnHolbein1 How dare you call my mountain of ongoing research projects half-baked!!

1dViews 419Likes 5
John B. Holbein@JohnHolbein1

@joefrancis505 some (not me) would say that all replication work is half-baked.

1dViews 164Likes 2
John B. Holbein@JohnHolbein1

@ryancbriggs all human, baby!

1dViews 428Likes 1
Ryan Briggs@ryancbriggs

@JohnHolbein1 Huh? It was a joke about how most PhDs like to start projects more than finish them?

1dViews 71Likes 1
joseph francis@joefrancis505

@JohnHolbein1 wait until you see my non-replication work!

1dViews 50Likes 2
Ivan Korolev@ivankorolev89

@JohnHolbein1 In my experience, it has been a complete opposite. AI has made finishing projects (and converting ideas into completed papers) much, much easier than ever before.

1dViews 97
David Tindall@dbtindall

@JohnHolbein1 Interesting, if true.

1dViews 85
samir huseynov@huseynovecon

my prior is, even before ai, starting project had always been relatively easier than finishing it. of course, the definition of “starting” can change depending on person or discipline. in my line of work, if study design discussions lead to some promising design you can mark it as first milestone is passed. i tried reaching to this milestone several times with ai, but it is too quick and pretentious to convince me the design it discusses with me holds sufficient merit. however, it helped me a lot to move a good design that was done independent of it to implementation stage (coding of experiments and data cleaning). so, for me, it is more useful in finishing projects. jtc, cost of starting project for me is mostly mental (reading lit, trying to find gap, etc). so hard money wise it is “costless” but mentally it is not

1dViews 2
Adam David Long@adamdavidlong

@JohnHolbein1 @mattBernius Trying to build up the “spirit planning” muscle — decide on Sunday “what are we going to **finish** this week.”

1dViews 1